Podcast Archive - StorageReview.com - Podcast #130: The Latest from Pure Storage – ’24 //Accelerate Highlights

Episode Date: July 5, 2024

Brian invited Peter Skovrup, VP of Product Management for Pure Storage, to continue their… The post Podcast #130: The Latest from Pure Storage – ’24 //Accelerate Highlights appeared first on StorageReview.com.

Hey everyone and welcome to another live podcast. We've got a great conversation based right off of the heels of Pure Storage, the advancements in their flash modules, the new SLAs, AI integrations, all sorts of new evergreen stuff, or in continuing enhancements. So there's a ton to talk about and I've got Peter Skavrup with me who he lets me Americanize his pronunciation of his last name. Well he can't stop me at this point because I've already done it But Peter and I chatted while we were out there in Vegas and wanted to bring forward this conversation with our audience Peter thanks for joining me. Appreciate you doing this Appreciate you having me Brian looking forward to this you say that now but until about 15 minutes ago I didn't know you live in Pittsburgh and asburgh and as a cincinnatian i find that more than moderately offensive

wow hey there's such a thing as the winning football team no no you had to go with winning okay well i have a uh i've got a good friend that uh works down in north carolina in tech and i i often say the same thing to him and he he holds up what is it the steelers have five rings i think he tries to show me that with his hand which uh is also makes me feel sad but uh we'll move on we uh we we probably won't talk a lot more football unless you want to get into euros i know you're a bit of a european football guy yeah and we can do that maybe towards the end so it doesn't go down the drain all i don't know this might be more entertaining than talking storage uh okay so we were at resorts world for accelerate again a great event to me you guys run a really tidy event, nice and tight, lots of good content.

But the energy there seems to be from your customers and partners, at least, and your employees, obviously. But your customers and partners always seems to be at a really high level. What do you attribute that to? Well, for our Accelerate, which is our user conference with a lot of partners, analysts, etc., we always try to bring a lot of new innovations. And so I think the fact that there's so many new things to think about and different ways to use the storage system makes it really interesting. And so it's not just the innovation of features, but actually the way that you deploy

storage and the way that you think about it as a platform. And so overall, I think content is really important to an audience. And I think we deliver on some of the interesting aspects. Well, I know, too, that you guys are real proud of your Net Promoter Score. And it's not something we often talk about because I think it's a little nebulous to anyone other than brands that track that. But your score is like an 82 or something, which in your category is extremely high.

Yeah, you're absolutely right. We're proud of it. We put a lot of work into keeping it high and continue to evolve. And so it's not a surprise that you see a lot of people talking about it because we make it front and center in how we operate every day. To us, it's very important. It's built into the culture is that we want satisfied customers. We want that high degree of satisfaction around our solutions, but also our people and our processes and so um it's um you you

basically when you talk nps we bleed orange and and so and it's it's a baked into culture it's important for us and um you know we're not afraid to talk about it yeah and for those that don't know the net promoter score nps is is, correct me if I'm wrong here, but it's the willingness of a customer to recommend you to somebody else. Is that close to accurate? Yeah, I would say that's a good representation of it. It's something externally audited, so it's not something we set. It's basically an external entity talking to the customers and understanding their willingness to recommend.

You know, it's funny because I was at Veeam's conference a couple of weeks ahead of yours, and they too are really proud of their NPS. And it's funny when I think about the vibe, the energy at these events at a Veeam show is very similar to a Pure event. A lot of excitement, a lot of practitioners there, and maybe that's kind of the difference between your shows where you guys have a lot of technical people present and man, yeah, they're, they're always jazz. I, of all the reading we do of all the trolling on Reddit, you know, looking for,

for stories and what people are talking about in the zeitgeist, no one ever is unhappy with, with their pure arrays or capabilities or performance really. I mean, they're the it's, it's one of the most enthusiastic groups of people we routinely see online. So, I mean, obviously that's a big mission for the company is to drive that forward and drive that fanatical customer. You know, they're almost zealots for Pure at a certain point, I think. Well, I mean, we're happy to see the results, right?

It doesn't come by itself. And so, you know, it's always great when you achieve some results around it and can maintain it. So we do it throughout the whole organization. We do it through our products. And so this is part of who we want to be it's our identity well that's good so the products then are really what's driving that enthusiasm right and we should talk about some of that and I didn't even really give you a proper introduction your focus is on the

core products and maybe explain to the audience what that means. Yeah, so I am responsible for the product team for FlashArray and FlashBlade. If you're familiar with the Pure portfolio, that makes up the core parts of the storage platform. And so basically the FlashArray is our scale up system and FlashBlade is our scale out system and operates generally as unified storage systems. And so you're, I know software is important but you're a bit of a hardware nerd then as part of this this job role right? Well so as you can imagine Pure does both hardware and software and so the feature functionality generally comes from the software itself.

But there are some things that makes the Pure hardware unique. And it comes down to some of the capabilities and functionality that comes with it. So if you think about it, and I know, Brian, you've heard this story, but basically we do everything non-disruptive, including upgrading and replacing the whole system. You can't necessarily do that without architecting the system from the very beginning to do that. That's part of what we do.

Other things that you will see and hear when it comes to hardware is density and there's a very good reason for us to be very focused on that. It impacts how our customers operate their data management. And so if you think about a density and being able to increase that without increasing the footprint, without requiring more power, more cooling or anything of that nature, your ESG is a significant portion of this. So density is not just a matter of wanting to be first with high densities. It actually has a real business impact and so if you are to

accelerate for anybody who is listening here or for you Brian, we introduced basically our 150 terabyte DFM which is our flash drives and so introducing that means that we have by far the highest density in the industry. But it also means that our systems just became that much more efficient. And so going from a progression of 50 terabytes, 75 terabytes, now 150, and even going further than that has a real impact. It's not just about density, it's not just a technology aspect, but it actually changes how you operate your business and what you can do with your storage system.

Well, I know the direct flash module, I was able to steal it out of Kaws' backpack for about six minutes is what I was able to get towards the wrap up on Thursday afternoon,-up on Thursday afternoon. Or, yes, Thursday afternoon. And we made a short little social video of it. We got some photos of it. And the feedback was universally, you know, like jaw-drop emoji. And we told everyone this was coming at last year's Accelerate.

So last year, it was the 75- terabyte module and the 150 was coming. And oh, by the way, we're working on 300. This year it was a pretty final version of a 150 in my hands and then KAWS had a 3D printed version of the 300. So you can see the progression coming along. But like I said, the feedback was quite amazing on that. And by the way, I should take a moment here. While we're doing this live for anyone in the audience, if you have questions, submit them.

We'll take them if they're any good and get them to Peter. So feel free to interact and join in with the conversation. And we'll make sure to get those questions answered for you about core products. Or I think Peter's ready for just about anything so let's including football or football apparently so we'll we'll cover that too but talk about then the the DFM why it's so important to pure and and maybe if we can start that conversation

By what is it? and what is it not? How it's like an SSD, but not really. Can we talk about some of those high level differentiators? Yeah, absolutely. So the DFM is fairly fundamental to the pure technology. And so it stems from an interest in being able to really optimize and utilize the flash technology within the platform itself. So if you think about it, a typical SSD drive has a lot of additional components built into it for a couple of reasons. First of all, there's wear leveling.

There's the ability to recover from failed cells by having additional capacity, there's an abstraction layer in terms of an interface, and in order to deal with all of that, you need some memory, you need some processing capacity on an SSD typically. And so all of that comes with complexity, it comes with a price tag. So what you end up doing is you treat each SSD as a localized storage pool and you operate them that way. But what if you could change that? What if you don't need flash translation and components for that? What if you can look at everything

globally for a storage system? What if you can truly treat it as a global storage pool and operate that way. So now you're doing rare leveling, not on a localized set of the storage, but actually across the full storage pool. You don't have DRAM for the purpose of operating the data locally. You can do that centrally and therefore optimize around that. And what if you take all of that complexity out of the drives and simply just manages the flash in software. Now effectively you've created what we call the DFM because everything is operated in software everything is controlled and managed that way but we operate at the system level, not at the individual drive level. And so there's some significant benefits for that. And as I said, complexity goes way down

with that reliability goes up. Having global management of the flash allows us to go and essentially get a much longer lifecycle out of the same flash components on the system itself. So there's some real benefits to taking this approach, which is why we build our own DFMs instead of using off-the-shelf SSDs and get some real value for our customers through the system and how it's managed. And so that's why we aren't just looking at hardware. We're looking at the combination of hardware and software. And so we aren't just hardware nerds. We are also software geeks. And the combination of it is really what makes up a flash or an inner flash plate.

Well, I think it's important to remember that when Pure came to market, it's been, what, 13 years? 10, 12, 13 years at this point? I mean, something like that. It's been a little years 10 12 13 years at this point i mean something like that it's been a little while right yeah yeah so when pure came out it was at a time where hard drives were still predominant in in the industry there were flash systems there were all these hybrid arrays i'm sure we can all remember uh tajile and and there were probably half a dozen star, uh, uh, oh gosh, what was the starboard storage? There were a number there that are not with us anymore that were hybrid. And then there were some all flash arrays that,

that were out at the, at a similar time, but a lot of them were using, at least in the early days, even client SATA drives with what, like X25M from Intel was one of the first packaged widely available SSDs. And those were what, 40, 80 gig. I mean, so when Pure was doing this and starting to conceive of these arrays and this investment in technology, it was a total different time. But it's really come to you over all of these years to have your own flash module and is now a major differentiator for you clearly, right? Yeah, for sure. I mean, so no technology is steady. And so Pure took the approach early on to manage and operate only on flash. And so think about it.

If you have a system that's got to be able to operate on spinning disk as well as flash and do hybrid versions of it, you end up with a complexity. But more importantly, your software stack is a compromise between the different technologies. By going after flash and flash only, Joe was able to optimize the software, the management stack of this

completely around this technology. Of course, flash has evolved. And so what may have started out at TLC turning into MLC, QLC technology, it has evolved with that. And so obviously, there's a big difference between managing TLC and QLC, but it's still flash. And so the benefit of the Pure system is the optimization and how we manage the flash itself, regardless of the type of flash it is. And so the flexibility is built into it with a very strong optimization on flash only. Yeah, yeah. You were talking early on about power and your organization's energy efficiency goals.

And that, if we look at the DFMs, it's not just a system issue. You're reducing power consumption at the DFM level, too, while also simplifying your design by having less DRAM, for instance. Can you talk a little bit about DRAM? Because we chatted on that at Accelerate, and I hadn't quite thought through it the same way you guys had in terms of where failures happen in a drive, where heat happens in a drive, and where some of the power consumption and expense happens.

A lot of that goes back to the DRAM. Yeah, I mean, so basically what a typical SSD does is that it receives memory and it organizes, it receives data, sorry, and organizes and tries to optimize for the way that the seller set up, which, you know, there's lots of different layouts of flash that would take place within a typical SSD and so the idea here is that instead of having the DRAM

instead of having the processing and putting that on the drive you centralize that and so you don't have to design the same way when you don't when you have a centralized set of DRAM. It simply comes as a part of the controllers, and what it takes away is the failure domains. So think about what fails in a computer if you just have a standard server. Is it memory? Is it CPU? What would it be? And so the fewer components, the simpler technology, simply just having a bus architecture into flies and eliminating that is a cost it's a power consumption and it's a very reliability aspect of it and so

that's that's the difference that dfm makes and so we see it we see it in terms of the reliability data that we measure on our systems compared to your typical ssds and it's significantly higher like 2 to 3x for the because we take that different approach. And so managing flash is critical in a flash based storage system and if you do it well you can really see some benefits out of it. So that's the difference between it. So think about instead of having DRAM on every one of you on let's say 40 SSDs or whatever you would have in the system

versus just having it in your controller where you're running software anyway. And so a significant difference in terms of the power consumption, significant difference in terms of reliability of it. Simple is a better design typically for the reliability and that's what we do with it. What you see is when you go from a 75 terabyte BFM to 150 terabyte, you're doubling the capacity, but you're almost using the same amount of power. That just means that you cut the power consumption in half by doubling the capacity, if you think about our capacity unit you're right. So walk through that a little further.

So if I'm doubling the drive size or the flash module size, there are obviously some benefits on a per terabyte per watt, especially metric. What does it do to let's go back to software to the software stack of purity when I've got to think about total data under management by a single system? Does that additional stress require, or that additional capacity require larger metadata tables or any more complexity or more difficulty for purity as we scale up these DFM sizes? So clearly, purity has to scale with it, which is why we do a non-disruptive software upgrade of purity, and it just continues to evolve as the system evolves.

So it's as simple as just hitting the button for an upgrade of the system, and now you are managing the metadata at the level needed for the drives in the system. So it's really built for this continuous evolution. And so upgrading the drives to double the capacity is not a replace the whole system. It's just upgrade the system with the components that you need and move from there. Well, you guys figure these DFMs have at least a 10-year life or you maybe you don't pitch it that way but i think we've talked about that where where you think the longevity on

these drives is is quite extensive you must have some original drives or the gen ones that are coming up pretty close to that that metric do you guys, I assume you're tracking drives over time. What do those AFRs look like for some of the early product? Yeah, I mean, I think we do look at some of the really early deployments. We have systems that were deployed 10 years ago that are still running, serving the same data. But obviously, the hardware has changed in the meantime. The software has changed.

This is part of the non-disruptive aspect of it. But when you start looking at it, you simply just upgrade and replace as needed. And so if the drive continues to run and don't have any problem, keep running with it. If you need more capacity, go ahead and upgrade the drives.

It's a fairly simple approach to things. You do what you need when you need it. This is part of the Evergreen program where we say everything is non-disruptive, where you can upgrade and continue to stay current if you want or you can stay where you're at. It's totally your choice based on your business needs. And so it's tailoring the system design, it's tailoring the business processes and the overall business concepts around it as well. So it's not just enough that the technology is there, it's also that the business approach to it just follows along. So we've got a question from JK on YouTube.

They're asking, are there any speed trade-offs with the larger DFMs? I think that's sort of the same logic where I was going. What are the challenges with managing effectively double the data from the previous largest drive? Anything from a performance standpoint or anything else that would be different with the larger DFMs? So overall, there's a bus architecture, of course, within the system, which is PCI based. And so the systems are designed with the different generations to allow for the bandwidth that

we anticipate. And so overall, we've designed the system backplane such that you can continue to grow and utilize the bandwidth that the drives can support. That said, if you end up deploying and keep doubling the size of the DFMs at some point, you're going to end up with the same bandwidth, but just more capacity on it. That's a configuration choice that you're making. You can continue to go with your 75 versus 150 versus 300. And so the overall performance of the system and the capacity can be designed independently if you want. Yeah, we should talk more about that too, because I know with modern workloads

and we can get into this, that there's a lot of demand for performance and you guys have some new stuff there and some SLA driven performance guarantees so we can get into that but the one thing that we've hit on a couple times I want to make sure to spend some time there is the non-disruptive updates because that's really important to how you guys see part of the the big value benefits of the pure arrays talk about that a little bit more and what that means. Because I think most of our audience is familiar with typical two controller SAN systems where you'd upgrade one controller, then update the other and then migrate workloads or one takes over.

I mean, there's a number of different ways to do it. Most are active active or active passive, or we just fail everything over to one. But talk a little bit about how your upgrade process is different than that and what that includes, because it's a little bit more than just software. There's a lot of other things going on. Yeah, so maybe we split it into a couple of different categories of aspects for it. So let's start with the technology and let's end with the business aspect of this. The technology is designed so that you can replace drives, you can replace controllers, you can replace any module within the array without taking a performance impact. And so you can actually go to the point that you can replace the chassis that the drives are in that's serving the data non-disruptively and upgrade to the next generation backplane

and therefore the next box, the next generation of the system. All of this can be done non-disruptively based on the combination of hardware and software. And so, yes, you have to design your hardware to be able to do these things, which is what's unique about Pure. Then you have to have the software to control the process and actually make this doable and simple

for customers to get there. And so one of the unique things from the upgrade perspective is that even if you take one controller out of a Pure system, you'd have no impact to the performance. And so you can upgrade controllers, you can upgrade the whole system without taking an impact at all. So these are things built into the hardware.

Things built into the software is about the process and how you simplify this. And so we've done lots of things to really simplify the experience for customers down to and including the ability to simply just do a single click on upgrade and it takes you through the whole process automated. And since you combine that with hardware where there's no impact of actually doing an upgrade at your busy time, you're not locked in. You have the flexibility of doing this upgrade when you want regardless of what the business is doing now on the business side of things evergreen programs is all about this subscription to innovation to always stay current and it comes in a

number of different forms and so it comes as capacity as a lace it comes in terms of performance as a lace it comes in terms of of different forms. And so it comes as capacity SLAs, it comes in terms of performance SLAs, it comes in terms of the ability to always stay current on your hardware. So there are different versions of that. You have to design the business processes, the business programs around it to really allow customers to effectively do this.

So one is we have the capability of upgrading, but how do you upgrade in a cost-effective way and always stay current with your technology in a data center? That's equally as important as the technology itself. So when you look at Evergreen, in particular our Evergreen One, which is our subscription program, you simply subscribe to an SLA and that's what you are utilizing for the system. We make sure that it delivers that SLA and that's what you are utilizing for the system. We make sure that it delivers that SLA and we can do that using all the technology in the background. Well, let's yeah. Okay. So you covered a lot of ground there and I want to make sure we dive in on a couple of these points that you've brought up. So for your customers that are

just buying hardware in a traditional model, they put it in their data center, like the one, our little lab behind me, and run it just like any other storage that they're used to, the Evergreen program for them lets them get controller upgrades, software upgrades. It's a regular cadence as Pure comes out with new hardware so that they can always be on a modern version of that, more or less. Is that a fair characterization? Yeah, that's correct.

Okay. So you get to upgrade on a regular basis of your hardware and make sure you're always current with the latest. And to be fair, that doesn't mean Pure sends a new array and you migrate data and you have to, you know, remap all your LUNs or whatever, and then retire the old one, it's quite literally pulling the controller, putting the new one in and you're you go right, maybe not quite that simple, but

not far from it. Oh, it's pretty close to that simple. Yeah. And so you literally just pull a controller out while the system is running and serving data, put the new controller in, let it start using the new controller, replace the first one, and off you go. There's your upgrade. And so has that changed for your customers the way they think about hardware investment? Because I think most of the enterprise is pretty used to a three or maybe at the extreme

a five- hardware life cycle, but that with Evergreen and Pure was one of the, was early money on this. And maybe even the first to do this at scale. Does it change the math for how customers think about, well, I'm not really buying a thing for three years. I can maybe budget and plan for this a little bit differently. And it's more of an ongoing investment in capacity and some performance at that point. Yeah, I mean, definitely when people are looking at this from a finance perspective and saying,

well, should I really think about my storage system over a 10-year period and what the cost of that is? The benefit of the Evergreen program is that you don't even need to go that long in terms of your timeframe. If you just think about it, if you refresh your controllers after three years, you're staying current on that, but it's a known established fixed cost upfront. And so it's very easy to understand as a user, as a customer of this, what it means for you and not just how simple it is, but actually what the impact is for your environment and in particular the cost. And so what customers really like is the clarity and the visibility

to the cost of these programs that it's all established upfront there's not like okay you're five years into it now your support contract is going to be that much more expensive or you are no longer supported in this system you have to migrate all your data to a different system there is no migration there is no moving data around it's always upgraded non-disruptively for the system while you're serving the data and the cost is well established through a known program upfront. So it makes the overall concept of upgrading and staying

current very very simple. Yeah well it's based on so back to the NPS and why people like Pure so much. I mean, this is pretty fundamental to that in that I buy the thing today, I start using it, and then I get more stuff as part of that deal. Later, more power, more capabilities over time. It's an evolving thing and not just a monolithic device that sits in the rack for three or five years and then gets sent back to e-waste or something, which, and I don't think you guys talk about this a lot, but keeping more stuff running longer is generally viewed as a net positive too to all of these environmental objectives and goals rather than chucking stuff at three years, even though it still works and it's still contributing value. That's a substantial advantage as well.

For sure. I mean, so there's e-waste part of it. There's the always refreshing things in the system that gets it to be faster and more capacity. And think about the overall ESG impact, not just the e-waste, the power consumption, and being able to just store more and more data within the same system with very limited impact on doing upgrades. The other Evergreen thing I want to talk about a little bit more is Evergreen One. One of my favorite pure hardware guys, Tago, left and went over to the as-a-service side of the house. I mean, he was a hardware nut through and through. And you got him somehow.

Someone got him to go over to Evergreen One and talk about all these SLAs. So there must be something cool happening there, or I don't think Tago would be there. But we chatted for a while, and the SLA concept and some of the storage as a service products you guys are offering are really interesting, but they're also different. It takes a little bit of time, I think, to wrap your head around where that value sits when I'm buying an SLA for capacity and performance. Can you talk a little bit more about Evergreen One? What types of customers are engaging there? And just what you're learning there as you're getting feedback, maybe even just from Accelerate

recently from customers looking at that type of consumption model? you're learning there as you're getting feedback, maybe even just from Accelerate recently, from customers looking at that type of consumption model. Yeah, so I think you're absolutely right. It takes a little bit to truly understand what it means to do storage by subscription or more important SLAs by subscription for your storage system. And so the idea is that it's much simpler to understand what you're consuming in terms of capacity and performance, and it's much more cloud-like as an operating model when you think about just your SLAs and what you need for your business, rather

than think about the storage system design and how to size it. And so it really changes the mindset and also the business modeling of it. If all you do is you look at what you need in the form of SLAs, and you can simply just choose those as your criteria, and then leave it to Pure to make sure we deliver those SLAs. It changes how you think about it. It changes how you think about it, it changes how you need to plan, but because it's a subscription, it's much simpler to change. So imagine a scenario where a customer had bought a system and all of a sudden their workload changed. They may acquire an organization,

they may all of a sudden have a new part of the organization that needs storage in a very different way. So both in terms of scaling up and scaling down of your environment, whether it's capacity or performance, now it's a subscription question and you can simply add to it for the purpose of solving those problems. And so again, you don't have the ownership is not on the customer of planning for these changes. That ownership is on Pure and we know how to scale systems. And so simply just looking at the SLA and that's what you're buying

and having the flexibility to change it over time makes it very interesting for customers to go down that path. And so we see a lot of customers switching to a subscription model instead of the traditional model. It is absolutely an educational aspect of it because it takes a little bit of time if you're used to building and sizing storage systems for the purpose of your applications all of a sudden just having to say what you need and then the system will deliver that it is a big change and and so not everybody uh catches on right away,

but it actually makes a huge difference in how you think about it. And so just think about this, you have a performance SLA, no matter what you're doing in your environment, we'll guarantee that we deliver what we said in terms of performance. And if you need to increase it,

we can increase it without taking any hit, without migrating any data, moving any data. If we run out of space on that controller, we'll just simply upgrade the controller as a part of your subscription. And so the flexibility and the dynamic aspect of running your business is now reflected in your storage infrastructure, but as a subscription. That's a fundamental change. Well, that only works for you, though, if you have already sorted out these other things like the non-disruptive upgrades, right?

Because if it's all difficult and cumbersome, then you wouldn't be able to offer it this way, or at least not quite as nimbly as you describe, right? Yeah, of course. I mean, why not take advantage of all the great technology? That you already have. One that we didn't talk about yet really is Pure One

and some of your other data visibility products and benefits you can deliver through that. Tell us a little bit more about that and what that layer is on top of what you do with just the standard Pure array. Yeah, so Pure One is our management as a service. Basically, it allows you to look across your whole organization

and see your full storage environment and effectively manage it that way. So Pure One has the benefits of not only seeing your complete environment as a customer, it actually sees every customer's environment. And so what that allows us to do is a lot of interesting aspects of it. So there are certain aspects of Pure One that makes it very simple. So let's say you're running a Oracle database or SQL database, you pick whatever application, and you now need to run another instance of that. You can simply use what we call our workload planner, which will simulate that workload based on your environment, as well as everything else we learned across thousands of customers on what this means for the array and what it means in terms of what's needed for this application and so being able to take

advantage of a lot of learned meaning ML and information that we have data gathering on how applications behave and what's needed from them really simplifies the experience for the customers. So being able to understand upfront what will it take to deploy these another 10 instances of my database and just letting the system tell you, here's what it's going to mean, here's what you need for it. And by the way, if you need that, click here to get in touch with your account team to

basically tell you what this means or for that matter go even further with an evergreen one and simply say I need to upgrade my subscription to this and so tons of benefits comes with a pure one we look at things of workload planning taking a lot of the headaches out of that on a pure platform we look at things like anomaly detection how do we go and actually learn a lot of the headaches out of that on a pure platform. We look at things like anomaly detection. How do we go and actually learn a lot about the environment? How do we get a highly credible way of looking at anomalies within your environment? Or how do we do a security assessment of your environment that allows you to really understand, are you as secure as you

ought to be within your environment? Do you have all your workloads protected? Lots of people don't have that full picture. It's a very manual operation to actually fully understand what's happening within a larger environment. So Pure One takes an approach to simplifying operating at scale, operating across fleets, and really understanding it as well as bringing in the experience of all the other customers and giving you that benefit. That's fundamentally what it does. Lots of interesting aspects of Pure One in terms of how it achieves it and what

the technology looks like. But if you just want to look at it fundamentally, it simplifies your operation at scale. It takes away headaches and friction in terms of operating and optimizing your storage environment. Yeah, I mean, I think a lot of the work that you're doing there, and we don't have to spend a ton of time on it, but Copilot was another technology you guys were showing at the keynote, which is a little aid, an LLM to help people with their questions or what volumes are using more resources or those types of diagnostics data. I mean, that was pretty cool.

What was the feedback you got on Copilot? It's very interesting, a lot of excitement around it. So what we did with Copilot, we introduced a lot of excitement around it. So what we did with Co-Pilot, we introduced a generation of Co-Pilot where we basically took some of the typical storage admin workloads or workflows and tried to automate them. So imagine that you could go in and say, okay, well, I have a set of workloads. I want to understand how they are operating. I want to understand whether they are optimized. I want to know which one of them may be putting

an undue load on the system because DB admin is deciding to reindex the whole database in the middle of the day or whatever it would be. It actually helps you clarify those things. And so lots of lessons learned on our side of the storage system and what it means to optimize them. It's built into the data set and what comes with the LLM. So the co-pilot is able to get you data that otherwise would be much more difficult to get a hold of and present it in a way that you get the view that you need in order to take actions and actually be able to optimize your system. So in some of the cases we showed, how do I find out what my busy applications are and how do I go and actually understand my security score as an overall fleet.

How can I actually get an indication of do I have consistency in my fleet? Do I know whether I'm operating at a reasonable security level for my install base? And so the LLM is able to basically give you those answers very quickly in a language that's much simpler to understand than necessarily diving through 10 spreadsheets and figuring out what it really means. And so it's utilizing the technology to just make it much simpler to gather the information, format the information, and deliver it in a way that you can consume it. And so Copilot is meant to do that and

will continue to evolve and bring in lots of use cases for the purpose of simplifying the storage admin role. And ultimately the goal is to be able to have more generic technologies be effective with a storage system. Well, I mean, fundamentally to take a task that might be, even if it's five minutes or 10 minutes and I get it with a couple of keystrokes instantly is, uh, is pretty big. But I think once you start combining these things and say, show me this and this and this, and then make it a list and then, you know, whatever, like it, it gets, uh, additive really, really quickly. I made a Tegile comment. Somebody, you've got a new customer watching us.

He said, currently replacing a Tegile with a new Pure, which I find actually probably says quite a bit for Tegile that it's still running this late in the game, but good for them, but new customer for you. So that's good. We've made it almost 40 minutes, and I'm going to make my first AI question, but it's not because I want to.

It's because Crimson on YouTube wants to know. There's a couple things here, so I want to break it down a little bit into a couple pieces. But the AI workloads and AI means so many things to so many different people. And I hate just saying AI in general, because it's not the same. Fueling a server with two GPUs in it, or even eight is a lot different than, you know, these rack scale needs where data has got to fuel these GPUs extremely quickly or millions of dollars doesn't get used. Where should we start with this?

You guys had NVIDIA at the event. You had a Q&A on stage. We met with them one-on-one after that. Tell me what's kind of the latest in the NVIDIA relationship with Pure and what that looks like. Yeah, so obviously there's lots of interest in high-performing storage systems and getting connected to NVIDIA GPUs. And so we're working very closely with them, have a strong relationship on where this is going. We've gone through a number of the certifications and we expect this to continue to evolve.

Pure systems, and in particular our Flashplate system, is extremely well set up for an AI scenario. So we think about it as HPC and AI being a part of that. And so when it comes to a GPU business and in general NVIDIA, there's a few things that customers are really interested in. First of all, they want to be able to connect. They want to know that there's a solution that works with NVIDIA. And effectively, that's what we're delivering

through some of the certification and solution programs. And so we announced that at Acceler there is a um a super part ethernet based super part certification coming in in collaboration with nvidia and so we think that's very important because in an enterprise environment and as ai goes broader in many different ways, Ethernet infrastructure is probably the future. And so it's in our mind, it's a matter of when, not a matter of if. And so we really wanna facilitate that. So that's probably the first part of it.

The second part of it. While you're there real quick on the SuperPOD bit, at scale, which is what you're talking about here, the biggest limiting factor for storage, well, I guess there's a couple, but one of them is the fabric itself. So on your flash blade, then what do you guys have to do on the hardware side to enable high-speed data flow? I mean, obviously, it's going to be quicker nicks on the back end, but do you, or is it? You tell me, what does that mean in terms of the networking side? Also, in general, when you talk about AI and high performance, et cetera,

you're typically looking at wanting to get to 400 gig ethernet as a step. And so that's very much in the cards for this. And so that's definitely part of it. Overall, the FlashBlade architecture is already built for this scalability, and so being able to scale performance and being able to scale capacity. So that's definitely part of what's happening with it as well. Overall, there's a different approach that comes into play as well. So you talked a little bit about it.

And so if you think about what your objective as a customer is and where the majority of your investment around AI goes, it's basically making sure you take advantage of the GPU that you invest in. GPUs are expensive. They make up the majority of the GPU that you invest in. GPUs are expensive. They make up the majority of the cost around an AI solution. And so you want to make sure that you keep them fully utilized for whatever AI task you're operating.

So what you really need to do is tailor the storage towards that utilization of the GPUs and making sure the storage system can serve that. And so one of the ways, and this is potentially a hard problem, right? So how much do I really need? How do I size for the GPU based on what I'm running and what I'm doing? And so we wanted to introduce some flexibility into this,

similar to what we've done with Evergreen. We essentially introduced a program for AI that says says we will deliver the performance that you need for your systems to keep your GPUs busy and so think about that as a subscription as an SLA instead of having to worry too much about the design of the system and so again it's a different way to think about it is that if you don't know what your future looks like with AI, and most people don't today, everybody is trying to really understand what they need and where they need to go. Having that flexibility and focus on the SLAs is another important part of how to think

about it. And so yes, there are things that has to happen in a system to get it certified. Faster NICs is generally what we're looking at. We can scale the performance of our systems to a very large extent with the FlashBlade architecture, and that's what people do for AI solutions. And so we build some really large AI implementations around the FlashBlade that's operating some of the fastest AI environments out there in the world. And that comes from being able to scale the performance as well as capacity, but mostly the performance

is what people are concerned about. And so FlashBlade has the option for that, and then tailoring it together with the SLA-based solution around it makes it interesting. What does FlashBlade do for you? And there's another comment on this topic from Crimson in terms of performance against other filers because it seems the unstructured is where most of the data is on the AI side, but a traditional dual controller rig may not be the best fit for fueling the GPUs as fast as they need to be.

So talk a little bit about FlashBlade and why you think that platform gives you such an advantage. Sure. So fundamentally, FlashBlade is a scale-out architecture. And the best way to describe it is as disaggregate storage. You effectively have a front-end performance aspect of it. This is the serving entity of the system. And then you have a back-end storage component of it.

And you essentially can scale these components for the purpose of designing to the solution that you're looking for so you'll insert blades into a flash blade system that then scales performance and the capacity and you can choose what those blades are for the purpose of the outcome you're looking for and so what it means is that you can start out with a relatively small system and just continue to scale it as your needs grow. And so it scales within a chassis and it scales the number of chassis in an overall single system without having to worry about the performance impact of whether you're in one chassis or the other. The system ensures the scalability of the performance of the capacity

so that it can grow to a very, very large capacity in a single namespace. And so that's effectively what you're looking for with the unstructured data, in particular in AI or more generically in some of the HPC environments. And so that's where the strengths of FlashBlade comes in, is that you have this

ability to, I would describe it as linearly scale the system performance and capacity, and you have the ability to tailor your system to your performance needs that you have around it. And you start out very small and just keep growing. And so we literally have chassis with a set of blades in them and we just keep adding chassis to the solution and that's how you scale it out. And so today it's 30 petabyte in a single namespace with the next generation of the DFM shortly coming out. We're then looking at 60 petabyte in a single namespace, and we'll continue to scale that. So we'll transform that into a name space. That's a good size namespace. That's not bad. It covers most people's needs.

Most, but I know you've got a couple of outliers where it definitely does not. So take that conversation of FlashBlade where you're going and talk to me about how it's different and where you think you have an advantage over some of the popular like parallel file systems that are out there. Because those guys have leaned in hard to AI to say, you know, they're the best solution for this. But why why is FlashBlade different than something like, you know, Weka or something like that? Yep. So, of course, AI still has many claims and it's very much evolving. I think a little bit of the the AI space still has being kind of like the Wild West, but nonetheless, it's getting under control. And so instead of necessarily comparing, I would call out what are the things that really matters when you're in an HPC space, when you're looking at AI. And so we've designed the FlashBlade

system around this ability to scale, but also the ability to perform. So when you look at unstructured data and particular files for the AI, at least at this point of time, it's about how quickly can you serve and how much bandwidth can you serve to ensure that the CPUs are utilized. What that means is that your metadata operations are sometimes as important, if not

more important, than your data serving. And so we built into the design an enormous capacity for dealing with metadata, which is critical in high performance files. And so that's where FlashBlade sets itself aside. It's not just the ability to serve and continue to scale it it's actually all the operations that are needed to effectively serve data that comes with it and so that's when the flash blade really stands out and we've seen it over and over in tests between us and competitive systems in customer environments and pocs where this is what really becomes important is what can you effectively do. And I don't think there's any systems really out there that beats the combination of scale

as well as scaling up metadata operations within a system. Yeah, it's an interesting take. And I think one of the challenges that the enterprise has with AI generally is as I make a larger and larger investment, you know, past just a couple GPUs, do I really want, how do I manage my data? Do I want a separate stack that just supplies the GPUs and if so, how do I manage data movement, data curation, in some cases, data backup and recovery? Lots of new challenges there. Based on what you guys were talking about with customers at Accelerate,

I'm sure this came up several dozen times. Are there any learnings that you that you have fresh off that event in terms of how your customers are thinking about this data support for AI? Well, I mean, as I mentioned, the program where we think about it as an SLA takes some of the headaches out of the customers today. Not knowing what your world looks like 12, 18, 24 months from now, just guessing at it is hard.

And so being able to rely on having a partnership with your vendor in the form of SLAs makes this simple. So taking an evergreen approach towards it to us, we get a lot of feedback from customers that this really resonates with them in terms of what they can do and how they can think about it. There are lots of things that comes with it, as you said. It needs to be possible to back it up. You need to be able to restore. You need to be able to do this at scale as well. And so it's a full set of storage problems that comes with it just at a whole different

level and not really knowing where you you're going to end up and so lots of investment in ai today lots of questions about what should i really do what what's my best practice around it and so this is where we are putting the story together not just in terms of the technology being able to guide customers on what what you can and should do and what your solutions that looks like, but also how do you actually protect yourself? How do you make sure that your investment is still valid 12 months from now

when things could be different? And so addressing the uncertainty together with guiding customers on what solutions look like and what works and what other people are doing makes a big difference. And so AI to us is actually something we've been in for a long time. We have some really large deployments that have been under the way for more than five years in terms of AI and significant learning environments. So we learned a lot from that.

We optimized our systems around it. That's where some of the technology that I'm describing is coming from. So it's not just guesswork, right? It's lessons learned over time working with customers that were front end of it. We use it for ourselves as well, right? So we don't own dog food in terms of this. We do learning internally on all the data that we have from the systems in the field. And for the last, I think it's eight years, we've had learning models built into Pure One that gets essentially where Purity gets the advantage of that learning that goes into it. So lots of lessons internally with customers are benefiting the new customers that are looking at AI.

So as big of a topic as AI is, not everyone is investing heavily there for a variety of perfectly valid business reasons. We've talked about a lot of the large stuff, the big DFMs, the huge namespaces available. Go the other way and talk to me a little bit about, from a core platform perspective, what does Pure scale down to? And where does it start to be interesting as an investment for a business? Is it a data footprint? Is it some sort of IOPS target? How would someone sort of self-qualify at the SME kind of range? Or you tell me what the starting point looks like for a small Pure, maybe it's distributed enterprise and they need them at the edge, but what's a small deployment look like there's a there's a couple of really good points here um so the the pure platform starts out actually fairly small

and um and is used by a lot of smaller medium enterprises and businesses overall um and so lots of customers have you know let's say something like 10 20 terabytes in a pure system and use it that way and go from there. So it's not that it's always these colossal systems that just have petabytes and petabytes of storage, right? It scales from fairly small and up to these very large scales of system. So part of what holds this together, and especially when you start looking at more of an edge setup or something distributed is where we start talking about fleets. And so, one of the things that we also have talked a lot about and did at Accelerate is pure fusion. And so, this is how we join different storage systems within the Pure portfolio together

with a single control plane to operate them as a fleet. And so in addition to Pure One, we've introduced the Pure Fusion, which allows you to build fleets of storage and allows you to really go think about storage as pools that grows rather than individual systems. So that could be a pool of small systems, it could be a mix of systems,

it can even be a mix of flash array and flash blade. So essentially, think about all of the storage technology as a single platform, where you're looking at storage as a service, you're looking at simplifying, managing this at scale at different locations. You want a simple infrastructure where you have the flexibility. You want to be able to upgrade and do everything non-disruptive.

Now you put this wrapper of pure fusion around it, and you can, in your brownfield world, essentially go and tie storage systems together and make them operate as a storage pool and so the ability to effectively go in and now say your data center or a subset of your data center is now an availability zone where you do non-disruptive operations and being able to operate pools of storage and utilize your infrastructure much better comes with that. And so it's not just a matter of the scale aspect of it, which we allow through the component choices within FlashArray and FlashBlade, but it's actually the utilization of it and how you optimize around that. So you could imagine that you had different models of systems in

your environment, you had different capacities. How do you actually optimize these systems so that you get the maximum benefit out of it? We have two answers to that. One is the fusion, which allows you to tie them together and really operate that way. If you were at Accelerate and you had a chance to see it, actually linking storage units together is a matter of like 20 or 30 seconds, and now they're operating as a pool, even with data in place. It's just phenomenal to be able to see that you all of a sudden create a cluster out of individual storage node, and there's nothing involved in it other than you now have data and you can think about it as a single pool. The other component of making this simple is through our evergreen programs where you can start rebalancing and you can actually think about your enterprise where just because you saw you were going to end up using a lot of capacity on one system versus another,

you can effectively redistribute your capacity at an evergreen level. So you're not locked in forever. And so those are aspects of scale of systems that come into play. I know you were asking about the small systems, which we do scale down to, but at the same time, most pure customers don't just have

one of the smallest systems. They end up with multiple systems, which we're very happy about, but we want to make that incredibly simple. And so that's the technologies we apply there. Yeah. And I think it's an important point. And this is where you and I started was that it's not just the hardware, even though we're know, I, you know, we're hardware guys to you, you are as well, but it's the whole ecosystem. And from going small to the largest AI deployments, all these software components are really, really compelling. And obviously what sets Pure aside, it's just more than an array vendor company. And I think that's the important takeaway

for people that don't know Pure. There's so much more under the covers. And we've hit on almost everything, I think, from Accelerate, that was the big news and the big updates. We've got coverage on all of those announcements on the website, storageofyou.com, and links to Pure,

and you can check out all that stuff. But this is a great conversation. I only regret we haven't had time or left enough time to talk about Euros at the end, but I know that you have some degree of sadness over the outcome for your team. But are you pulling for anybody that remains? I kind of lost a little bit of the attachment

once your team was out, right? So, you know, you can't switch to the enemy overnight. No, but, you know, so the U.S. is about to get punted from Copa America, it looks like, but U.S. soccer fans take great joy in watching Mexico get booted, for instance. So maybe you're rooting against somebody then. You know, right now, my time is going on the European Cup.

Sorry. All right. Well, you know, there's always next time. Well, this has been awesome, Peter. Really appreciate your time and joining us on the pod today. And again, Accelerate was a great event. Pleasure to meet all the pure people there. has been awesome uh peter really appreciate your time and uh and joining us on the pod today and again accelerate was a great event pleasure to meet all the pure people there yourself included and and we were happy to be a part of it again this year and uh thanks for your time thanks for

doing the pod yeah thanks for having me appreciate it very good