00:00
Now, so welcome to remind me, Matt, we are at the next generation scale up architecture with Flash array, man, that's a fun mouthful. Or if you saw an announcement side that said Flash ray XLR 5 and you're like, I want to learn more about that. Hey, you're in the right session. So this is gonna be a little bit of a different
00:19
style session, hopefully a little bit more relaxed. Uh, my name is Andrew Miller, lead principal technologist. I Introduce ourselves and the site if we haven't actually if you I've actually met you in the past or Moni or Matt, please come up and chat with us afterwards. Questions throughout. We do actually regularly host a coffee break
00:33
webinar series. That's part of why I'm here, where we actually have more kind of like podcast and webinar, and Matt and Mike and all have worked together enough and like or dislike each other enough that we might make fun of each other in front of you all the time, and we'll see if I keep my voice through out. So with that,
00:47
let's dive in a little bit, uh, introductions, like I said. Andrew Miller, lead principal technologist, Pure Storage. I've been here about 6 years now, uh, customer side for 8 years, partner side for 7 years, built out a technical marketing team. I try to cover the whole pure portfolio, man, that's fun and hard and challenging.
01:02
I work with deep folks in various areas and host various external public facing things. So if I keep my voice, I'm here to play a little bit of moderator and facilitator as well as on some of these. I'll actually talk about some of the advancements we're doing on the file side. I love though being joined by the, the orange flower in the middle.
01:19
Can I say that, Matt? Matt, please introduce yourself and M. Hi guys, uh, Matt Hallberg. I've been at Pure for. For 2.5 years I've had the pleasure of uh launching XCR 4 in 2023. Last year we had a pretty fun live NDU discussion with some of the financial tech companies. Um, excited to be here with you guys this year.
01:39
It's been a pretty amazing show and with that I'm gonna pass it to my old boss. Oh, OK. Hi, hi everyone, uh, my own Patnagar. uh, I lead hardware product management for Pure Flash, Flash Blade. This session, although it is about scale up on flash ray, here to hang out if there's more questions on any part of the portfolio.
01:58
Welcome to answer. 7 years and running with Pure, kind of seen the portfolio grow with Pure and get into various parts of our customers' environments, so really excited to kind of bring more of the recent announcements and exciting news over here today. And you wanna remember Flash AC when it launched about 5 years ago.
02:16
So Mike and I started first started working together. It was a lot of fun, so we get to do it again today. OK, diving in a little bit. So thinking about, let's set the framing a little bit. You know that you're in a flash array XL session.
02:29
That is it. Well, until we talked about this cool thing called Flash ray ST on stage, more about that, but we're in the world we're in the world of relatively large frames, very high performance. We're thinking about the most critical applications out there and potentially really large environments, unless you want to have just one giant flash ray XL to roll them all,
02:47
that's fine too. So when we're thinking about this, we're thinking about in the frame of questions like as I'm expanding my data center, continuing to. How can I add more without physically expanding? Is that possible? sometimes not, sometimes it is because when we started to think about physical expansion,
03:03
data center space, power, cooling, that's actually a lot of work, right? How can I manage all of this as I keep growing, do I need more people? Who here gets given more staffing as your environment grows reliably? Anybody? Probably not, right?
03:19
I'm seeing a couple, a couple of head nods or head shakes. As well, ESG matters if you're from an EMIA standpoint, but even within the US, where sometimes we see that more of the global standpoint, the sustainability goals, still the level of power that's consumed is a big deal. And then last but not least, technical debt is real.
03:38
And that I keep expanding my environment and I incurring more and more technical debt as I bring things onto the floor. I'm bringing things in today, then in 3 years I know like, mm, I'm gonna, I'm gonna pay the price for that later or can I get ahead of some of those problems or challenges? So we've embraced these questions in various ways from a flash rate standpoint.
03:58
And where we wanted to start today was actually with the core architectural principles that haven't changed as it relates to flash rigs. There are things that have changed, but actually both kind of reassuring you but also reviewing a little bit the core pieces that we're still building on and still holding true to. Would you mind taking us. a little bit mild.
04:17
Absolutely. Now to kind of make it a little interactive, just a quick show of hands of how many are kind of very well familiar with Pure quick show of hands, like pure customer, pretty familiar with Pure perfect, completely new to PR evaluating PR. Thank you. You've got a lot of people to answer your
04:38
questions. That's the summary of it, right? So, ah, anyhow, yeah, so I think, ah, you know, a lot of what you brought up as as questions or challenges, ah, essentially. In our philosophy, in our architecture, a lot of these things are kind of intentional in our design, in our architecture as to like how do we kind of think out of the box and as we are
05:01
solving customers' problems, like how do we make sure it's done in a much more genuine and bottoms up manner instead of like it being an afterthought. Now I won't go in this exact order in the slides out here, uh, but essentially a few key themes that you'll see. One which is kind of, you know, centric to most of our announcements that you'll see today in
05:24
this session, keynotes, etc. is the concept of doing more with less. Um, it's a simple concept, but that's a pretty big design philosophy and ethos, and I, I wanna mention these architectural concepts now. But as we get deeper, we're going to give some proof points, right, not keep it as fluff, get into real proof points of like how did we deliver that,
05:46
right? So doing with more with less, whether it's like storage efficiency, whether it's performance efficiency on a single rack unit basis, those are kind of key tenets of our design principles that's, that remains, that continues to remain, and you should kind of hold us accountable if you don't see that in the future.
06:05
So that's one. Now that's has a pretty good and strong connection with total cost of ownership. Obviously if I'm designing for doing more with less with every release of ours, if we focus on metrics like performance per rack unit, performance delivered in the most efficient manner over to our customers for their
06:25
applications, best storage density that kind of results in a smaller footprint. Uh, lower power consumption, lower space, not just that, like as your environment scales, it translates into like easier operations, fewer people, uh, to run those operations to what you called out. It doesn't stop there, right? It extends into the last pillar over here which
06:49
is shown as forever and that's again, you know, forever is a marketing term, but there's a lot underneath it. I describe it in a nutshell as. The forever concept or the business concept was again not an accident, was only made possible just because of the design thinking. The forever mindset is, hey, if you're a pure customer, you're a customer forever in the
07:10
sense that none of your technology goes obsolete. Now, competitors can kind of mimic that on the commercial terms, right? Like, hey, you know, if you're under support, none of your technology ever has an end of support date. But that would start turning into a massive expense if you didn't do it right.
07:28
The underlying ethos and philosophy over there is we've got to design components that stay much longer out in the field, outpacing any of our competitive technologies. The key component that resides on a storage array, give a simple example, the Direct flash modules run with our Direct flash software as an example. Our annual return rates of that are 0.3%.
07:53
That's 10 0.3%. We measure that on a quarterly basis, on a regular basis, that's 10x, then most of the industry standards that you see published out there for whether it's SSTs or HUDDs, but that's just a simple example of like if we don't. Engineered these things hard enough.
08:12
Some of the commercial, um, you know, motions that we have like supporting a forever uh support program will never be viable. We design chassis so that they live at least a decade out there like the current flash array chassis that we have. Matt's going to introduce something new, so I want Steelers Thunder that's lived for the last
08:32
10 years. It was designed intentionally so that we could live 3 generations to 4 generations of 30% performance improvements through that vehicle. So no, no, none of these are by accident. These are core tenets of our of what we designed. Uh, now, simplicity at, at a single year unit, great, check the box.
08:51
Now, as our customers grow their environments. Simplicity at scale is something I want to kind of ask you about, which is, um, you know, you heard a lot about the intelligent control plane, a little bit of more flavor, uh, Andrew, from your end. So in one second we'll talk about the building blocks, but continuing to think,
09:09
thinking about scale, there's scale from a standpoint of it at a human level and at a management level, but it can be something changing how we want to manage things. So you've already heard on the main stage about fusion, right? So we can actually tell it's about. Fleet management, but there's a piece there of fleet management as you tease out some of the words,
09:26
not too many words on the slide you see simplified resource discovery provisioning, management about one logical data plane. Now I, I can't help but, uh, a few years in the industry and some PTSD saying we have looked at the attempts of how this has been done in the past. And while we are driving towards some similar goals, we've learned from the lessons where
09:47
there weren't common APIs where you had to put something in band from a data management so Fusion is very intentionally. Baked into the array, it does not create new resiliency challenges, but it allows you to manage at scale. And even like you heard Chad say this morning, I was waiting for him to say Chad Kenny on
10:03
stage on the main stage. He likes to say compliance from day one. If instead of deploying individually, we're deploying to presets or profiles, we're actually being compliant and what we need from day one. We even saw a customer in England. That actually for regulatory reasons is actually embracing this sooner than they
10:20
thought because they had a lot of different admins that were taking different approaches. It was actually a relative regulatory risk now. There is, there are much deeper sessions on fusion, but we might make sure to mention this in here because from the context of a flash array XL, think about large environments, the hardware and the software are critical,
10:38
but thinking about some of the management and like it says one logical data plane is a critical piece of this from a pure strategy standpoint, but even how you can handle this on a day to day basis. So, let's not hold the lead anymore. Do you mind, um, as we're thinking, as we're thinking about the building blocks of the most
10:56
recent Xcel, it's R5, right? Did we skip 23, and 4 minutes? Let me start with that explanation and then, uh, take us through the building blocks of more performance. Absolutely. So first off, very excited to introduce our fastest ever built scale up architecture, flash array Excel.
11:15
Uh, this is called Generation R5. There's a story to it. I'll come to it. Essentially this is about marketing and branding your your products consistently with the CPU generation across the board. So there's a little bit of, of, you know, our marketing and product managers battling what should we call it out. So yes, it results in us skipping generations
11:36
in between, but it is our latest generation of our of our most performing scale up architecture with all the rich data services. Um, what you saw on the keynote that Rob kind of pointed to is this is massive performance improvement compared to our previous generation, like a 70% improvement, 70% improvement compared to our previous generation in terms of performance.
12:01
By the way, we are being modest about it. When we say 70%, we are not claiming things like 100% reads off for cash. We want to keep it real. We want to keep it practical. These are 70 to 30 mixed workloads of a standard IO set. So we're not, we're not cherry picking is what you're saying.
12:17
You're not cherry picking, you know, there's a spread across in instances you can expect up to 2x the performance improvement that you saw in the previous generation. So what's fueling some of the performance improvements? So there's some that fall in the architectural speeds and feats. We are moving to the latest generation of the Intel CPU with Amerral Rapids.
12:37
Uh, we are updating our interconnects internally to be PCIGen 4. By the way, some of them are also PCI Gen 5 compatible. So again, the future mindset of thinking through like, yes, this is Gen 4 right now, PCIGen 5, whether it's connectivity to our direct flash modules, to hosts, uh, all that's already thought through, moving to DDR 5 with this architecture,
13:00
but that's kind of the under the hood scenes of like how did we accomplish some of those performance gains. One thing that I can guarantee and promise is expect to see this repeated time and again with the highest in order line of our products like Flash ray Excel constantly on the drawing board so that how can we push the envelope of some of these highest performance boxes that we have at at at Pure.
13:23
Um, now, as we do that, as you know, interconnecting connectivity, uh, improves across the networks and our infrastructure, this is going to be capable to run at 200 gigabit Ethernet speeds. So the network adapters, etc. that are with XLR5 also see a bump in in offering. Now that's more on the speeds and feed side.
13:45
I want to kind of again point back to the original uh challenges and design thinking, one being, obviously, you know, again, in a classic evergreen manner you can nondisruptively upgrade your flash array Xcels or even your X's over to this, uh, solution, uh, without any downtime. Like it's part of our evergreen story. Doing more with less.
14:07
Now we with every release we learn to do more out of the same infrastructure or less. We with Flash ray Excel, we're increasing the limits of our maximum raw capacity limits on Flash ray XL 170 or 130 by 50%. So we can host up to 2 terabytes of raw capacity. Most of these arrays, uh, see about an average of 5 to 1 data reduction,
14:31
so a 50% improvements in maximums is a significant multiplier if you multiply that with the the data reduction rates that you see. A concept that we introduced a couple of years ago was introducing certain offloads of functions on the array. We'd introduced the direct compression accelerator.
14:50
Now just to keep up to speed with the raw performance improvements, we've had to introduce a second generation of the direct compression accelerator. To just keep up with the ingest and throughput improvements, we now have DCA version 2. Um, which kind of offers against the enhanced compression rates that we introduced originally with XLR 5.
15:11
There's, there's actually one way I think that there's like two lensess. One is I think I saw almost all pure customers in here. So to some degree it's like this is just what you would expect from Evergreen. But that's really amazing at the same time too. You could have a nondisruptive upgrade that's included with Evergreen forever. And day one,
15:29
you don't have to wait for it, right? Perfect. OK, so getting into the applications and the use cases again, our portfolio's grown and evolved. You've seen on the scale up side our portfolio evolved from like Flash ray Excel all the way down to Flash array E depending on the performance, uh, propensity of those applications.
15:49
Flash array Excel. specifically is designed for mission critical database OLTP ERP like applications. That's where the key focus is, and I kind of wanted to kind of point towards again the performance improvements that we are seeing and you can expect with Flash ray Excel. The clicker is.
16:10
Acting up. I was going to be your clicker, but thank you, thank you. So what does that 70% performance improvement mean? Again, when we run benchmarks like TPCC, uh, for SQL server, what we're seeing is again that improvement in performance density. When we measure and benchmark against some of the competitive multi-engine offerings out
16:32
there, you get 3x better IOPs per rack unit. So you can stand up racks and racks to get to the same performance levels, or you can fit that in a 5 area, right? That's what we mean by delivering performance efficiencies where you can deploy massive performance in a much more efficient manner.
16:51
If you're looking at OLTP transactions, we are expecting 2 X to 3 x times new orders and the benchmarks that we've run specifically on OLTP databases. Again, like that's the ethos and the philosophy, we wanna, we wanna keep pushing the envelope on Flash array Excel. Uh, you should expect this with every release of ours,
17:13
but we're not stopping there. Also wanted to introduce something that's beyond the performance of Flash ray Excel called FAST. Rob Lee or CTO, his favorite, favorite product name product name so far this morning, um, now we also went into the drawing board to optimize anything we could to improve the
17:34
performance further and to introduce that product. Matt, I'm gonna pass it over to you. All right, thanks, Mike, um. Before we talk about ST though, uh, I wanted to see if anyone, I didn't see enough Jaws drop when we mentioned 70% or the 2X transitions per minute,
17:51
so I wanted to make sure that, uh, our wonderful crowd had a chance to ask any questions before we jump to the next slide, uh, because this is a forum and you guys can ask the questions that you want. This isn't a keynote, so does anyone have any questions before we move forward in the slides? And, and don't be shy. Yes, we got one in the back.
18:08
I knew it was gonna be Mark in the back, so. Yeah, start shouting it out and then I'll I'll get you the mic. Yeah I think you've got a point. 77 So, uh, I think again, right, I think kind of to serve the different performance needs of an environment.
18:32
did everybody hear the question? All right. So again, we're trying to create a a spectrum of products that can serve different performance needs and yes, the, the choice of the underneath resources such as CPU memory, etc. are kind of designed so that I can kind of hit
18:52
decent sweet spots between what we are seeing as performance needs. So the design targets are kind of different, right? Essentially the XcL 170 is where we're gonna continue to unleash the most, uh, Excel 130 again, like I said, those were averages. I do have data points where you're above the average up to 50% as well,
19:13
but Xcel 130, there are certain differences in the two products. One is about, I believe, 76 scores of CPU versus another being. Um, closer to, I'd say 84 if I'm not mistaken, or 6, 78 and 86, so those differences definitely kind of play somewhat a role in the ultimate, uh, performance gains, right? But again, like if you rank and stack our own
19:37
portfolio, um, you know, that is still about a 50% improvement over the X90s as an example. Like it kind of depends on the compare, if you may, yeah. Uh, good for you, validation. Anyone else? We'll be around afterwards, so yeah, if you're feeling shy. All right buddy.
20:02
So let me, let me point my phaser out here and see if it works. All right, flash your AST. So, uh, many of you stopped by the hardware pavilion yesterday, saw looks like a sheet was on top of something. The sheet has been pulled off today. Uh, there's a guy named Jason Sarek who is over at the hardware pavilion that's available to
20:20
talk about more details and what we have up top. But as mentioned, as my. was mentioning, um, we looked at our most performance platform and we said well what can we do to make it better? We actually have some performers, uh, some, some, uh,
20:36
customers that said I need to go to 11, right? I'm stuck on 10 with my XL 1170. I need to go 11 because I am all over, uh, the need for like super high transfers, super low latency, and while the XL 170 is amazing. I just need something a little bit more so we decided to release this product called Flasher AST again it's leveraging the Flash A XL architecture um as Mayan mentioned,
21:06
we, we took a very singular focus on performance and performance and latency where we did data path optimizations where now we're like sub 5 milliseconds, which is insane, uh, for a software IO path we got rid of a lot of the latency inducing. Features so really we just pared down this thing to let's say a poetic interpretation of performance. Uh, the enterprise features that you know and
21:32
love are still there, so we have snapshots and replication for data management or data protection. We have rideable clones, so it's, it's still in its soul is an Excel, but it's an Excel turn to 11 if you guys get the reference, if not, come talk to me afterwards. um, so here's actually a bonus sign we talk about.
21:51
OK, this is a spinal tap reference. Am I, am I being too obvious? How many folks know that there's a spinal tap reference literally in pure code? Does anyone know about active cluster or synchronous replication? It can handle up to 11 milliseconds of latency.
22:07
Your application may need less latency than that. That's fine. That is a spinal tap reference in pure code. Thank you. All right. All right. And so micro milliseconds. Sorry, did I say microseconds or did I say microseconds here,
22:21
but we're talking replication and milliseconds. So actually sorry there's one other thing here I forgot. So, so if you're, if you're a partner in here and you're thinking like who would this be for? you're a customer, the way that I. Sometimes express this when I'm chatting with folks that if you say, hey, some millisecond, 250 to 500 microseconds,
22:39
that's too slow for me. And I know for every 1025, 50 microseconds what that means in additional revenue generated for my business. And I'm not actually asking the price tag on that. That's what this is built for. No, that is not saying that the price tag is so
22:55
high, but it's actually built for folks who are saying like I can actually actively monetize that decrease in latency tying it back to my revenue generating sources. Yeah, just a quick note to add this, you know, some of you may have heard of this product, uh, to various briefings, etc. you know, originally when we, or even now this product can kind of serve up to.
23:15
Database applications of the size of let's say 100 terabytes, but soon enough this is going to expand to 400 terabytes. So originally when we got a lot of feedback we're like, hey, this may be too small for my needs, but that barrier is going away pretty quickly as well. So it's kind of becoming way more compelling in
23:32
terms of posting reasonable size, uh, databases that really need quick transactions and a good way that, you know, Rob described it on the keynote stage was like. If you're looking for a DAS replacement, this kind of gives you that it's in the ballpark of the, the latency that you get from direct attached, but it gives you the benefits of all the uh data service capabilities that exist like snapshot management,
23:58
replication capabilities, encryption address, etc. So that's another value problem that we've seen customers uh latch on to. Oh, all right. All right, and so for the products that I personally love and enjoy the most because I everyone has a a favorite child, don't deny it.
24:17
But what's new for Flash array X? Uh, so one of the purposes of, of, uh, this, this session is to talk about some of the things that have happened in the last year since the last Accelerate. One of the cool things that we released earlier this year was, as Maya mentioned, our next generation chassis.
24:34
So, um, as you guys might be aware, like I think Chassis Gen 1 came out with our M series 10+ years ago, and we've been able to. To to use that chassis in all of our NDUs right, going from M series to MR2 to those who did X7DR1s to XR2s to XR3s to XR4s so we took a look at um the the chassis and we said, you know, it's it's time for a refresh. We're gonna do two things one,
25:02
and I talked about this two years ago we, uh, introduced this concept actually with Excel when. Excel first came out they use these drives called DFMDs. This was our, our pure zone direct flash modules, but we put the uh NVRAM actually on the drives themselves. So if you were to look at an Excel, there's no NV RAM drive slots,
25:21
and for those of you who have an X or a C, um, you would notice at the very top of your chassis you have these 4, soapbox looking type things. Those are the NV RAM drives that are on the top. With, with Excel, we took the MV RAM that was in those drives and we moved it on to the drive itself, and then we were able to curate a right group across that MVRAM.
25:44
Eliminating the need for for NVRAM drives specifically. So, uh, well, what are we gonna do with the next generation chassis? Well, let's leverage the DMVR on the drives. Let's take those those drive slots that were for NV RAM and let's convert them into actual drive slots that you could uh you could plug drives into.
26:02
So what did that result in? Uh, 40% more storage, right? So instead of 20 drives in a chassis, now we have 28 drives in a chassis, and that, uh, has been cascaded down to X to C to E, and I'll go over the impacts to C and E next, but for X, what that was able to get to us is we're able to drive with the 36 terabyte drives,
26:24
uh, 1 petabyte in 3 rack units. So it doesn't seem like that big of a deal, but 1 petabyte in 3 Iraq units is is pretty crazy from a capacity standpoint. Uh, and you know, with that one petabyte, I just want to make sure anyone who's an X90 user out there, um, we actually gave you guys the ability to go to 1 petabyte in your own systems
26:45
as well. So it's not 1 petabyte if you buy a chassis gen 2, you can take your existing X90 R2, R3, R4 today and actually expand the capacity up. To 1 petabyte um and that's gonna be purity dependent so you wanna be in at least 6.8 when you do that. So software upgrade increases your maximum capacity. Oh yeah, yep, and again it's that concept of
27:08
being able to do more with less. It's a concept that everyone says they do. No one actually puts into practice except for us. We actually put it into practice. All right, with flash ray C, um, and before you go to that one, can you elaborate on the underline bullet points for you purchase only.
27:28
So there is a specific marketing SKU for Chassis Gen 2. So if you're working with your partners or um you're working with the ASCs and AEs, there are specific marketing SKUs today for X70, X90, C70, C90 that have a CG2 in the marketing SKU name that basically signify they come with chassis en 2. It'll come with this 28, uh, chassis system.
27:54
Um, we're, we're covering those models specifically. We wanted to do a limited launch on this ahead of any other product generations that may be coming out. Uh, so we wanted to, to, to really just limit this at the time of release to the 70s and 90s to really showcase the performance benefits that you get out of having a, a Gen 2 chassis, as I mentioned,
28:14
um. So there was the 28 drives instead of 20. The other cool thing that we did this is, is we wired up the back plane for Gen 4 and all of our drives, uh, all of our drives recently produced since August 2023, we call them Gen 4 drives because they're PCIA Gen 4. So you're able to extract the PCIA Gen 4 performance out of the drive with this new
28:36
second gen chassis, um, and that's why it says for new purchases only so. If you buy a new system you can get it with this second gen chassis for the for those of you with an X90 or a C or something else today and you want to take advantage of this, we will have the ability to chassis ND uses in the future. It's about a year out, um, but it is something that we will be offering to our customers who
28:58
want to take advantage of all the cool things that come with chassis. We actually have a track record of that, although it goes back a little while from which introduced the very first X chassis on the M. We didn't have a chassis before, so we've actually done non-disruptive between chassis or the X of the XL so it's hard, but we have a track record of doing it.
29:17
Yeah and with when you say it's hard, I mean that's one of the cool things is um I've been I've had the pleasure of working with their engineering team over the next year and why it's a year out is we are looking at addressing some of the, I wouldn't say they're complicated things with doing a chassis in you but we're greatly trying to simplify the process even more than how simple it is today.
29:39
So, all right, uh, flasher AC there we go. Again, same principles with Chassis Gen 2. The part that I'm like crazy excited about, as you guys know, we released the 75 terabyte drives, uh, what, 2 years ago. Um, one of the cool things that you get with moving to Chassis en 2 is Used to have this
30:01
capacity max at 2.5 petabytes. Now you're at 3.6, so 3.6 petabytes for your business critical workloads, uh, for your cyber resiliency, for your snapshots, for your VMs, all that kind of stuff, 3.6 petabytes, and again, if you have a C90 today, you can upgrade to 3.6 petabytes by adding more drives to your existing system or adding a
30:25
shelf. The fun part is with those 75 terabyte drives and this chassis, now we're at 2 petabytes and 3 rack units. So we have 1 petabyte on the X. We have 2.1 petabytes on the C and then we get to the really, really exciting part is flash array E. So as you guys know, we did 75 terabytes, uh, two years ago,
30:50
we announced 150 terabytes last year. We started shipping 150 terabytes uh earlier this year. Again, to reiterate, 28 drives in the chassis, 150 terabyte drives. That is 4.2 petabytes in 3 rack units. 1200 watts is our is our typical power.
31:14
1200 watts. 4 petabytes, 3 rack units, and it's, I think, 4200 BTUs. I, I used to say less than the toaster. You've been a pure customer for a while. I think we'll update that like less than an air fryer now. That's maybe newer and cooler, but it's almost a scary amount of space capacity in that space.
31:34
It's insane. Like I, uh, I remember seeing some reviews about this one system that was like, hey, it's, it's, uh, it's 5 rack units, it's about 800 pounds, it fits 108 3.5 inch drives from Seagate, and it gets to this amount of capacity. I think it was like 1 petabyte of capacity. But you have to use a forklift because the thing is heavy and you can only install it on
31:58
the bottom of the rack. You can't put 2 in a rack because you bow the rack. So if you want 2 of these things, you have to have 2 racks and then you put the other stuff on top, literally 100 pounds with 28 drives. you know, 1400 watts. I just, to me this is an insane concept.
32:18
I've been in the drive market industry for a long time, so like the idea of a 150 terabyte DFMs, the idea of being able to shove 4 petabytes in a 3 rack unit system is just like mind blowing. All right, since I'm, but there's more. Oh, there's more, yeah, there's always more.
32:36
Now since we're talking about DFMs. Let's talk about 300 terabytes. Yes, we expected, I think we, we missed planned a little bit. We expected that cosms aren't going to be on stage waving around a 300 terabyte DFM. So pretend that you've already seen him doing that tomorrow as you keep talking here.
32:55
So have that image in your head of Cos. Yeah, and I, I did verify. If you guys see him walking around, ask him if he'll bring it out of his backpack or his pocket. All right, I did verify it was the first thing I saw when I saw cars. I didn't even say hi.
33:09
I said, do you have the 300 terabyte drive? And he's like, Yes I do. So all right, 300 terabytes is real. We announced this last year, right? Cars was up on stage saying we're gonna have 300 terabytes in a year. We have 300 terabytes coming by the end of the year.
33:27
We're gonna be focused in on uh our E family, but if you think about the metrics that I showed you before, like for flash array E, dude, 28 drives. Now we can't handle. Uh, 28 drives at 300 in the FEE right now, but we can in the future because that's a lot of capacity. So, um, again, you know, the, the best thing
33:48
about direct flash modules and QLC and us handling it is we're able to scale to 300 terabytes. I actually saw a storage review article, uh. This morning that was a review of uh a competitor's 122 terabyte drive that just came to market and I was like that's cute 122 terabytes we're already shipping 150 and we've got 300 coming in like 3 months so.
34:14
That's that. Obviously I'm super excited. Uh, so, and the cool thing about the 300 terabyte drives is again, um, you're able to do capacity consolidation. So in the future, let's say if you have a a flash array E or a flash blade E and you want to again take your existing footprint and dial it down, let's say that you have a flash ray E with a shelf and
34:34
you really want to get rid of that other 3 rack unit shelf, you could do a capacity consolidation down to again that 3 rack unit footprint. So with that, uh, at a, at an observation there, right, which is like we've been, we've been geeking out about all the hardware stuff, but there's stuff that probably slides through without people noticing, and I want to give an example as we switch gears,
34:58
which is. Uh, we make purity upgrades all the time. Keep an eye out for the release notes. So I want to take the example of, uh, E family. A year ago, we just sneaked in a code update with said like, you'll get 5% more capacity with the same implementation.
35:14
Just do the purity upgrade. Now, let's put that in perspective of multiple terabytes. 5% is an easy 150 terabytes unlocked for free. Right. So that is us kind of making sure we can constantly evaluate our backend implementation of FMM and Purity,
35:35
or direct flash implementation and optimize it and look at it per platform, right? Like, for example, design choices we may have made at X. Which requires a certain amount of performance profile may not be relevant for a product serving a different application like E uh does, and we'll keep unlocking that.
35:56
I think in a few months we'll be unlocking more usable capacity out with just a purity upgrade. That is just a constant evaluation of like, hey, did we have any inefficiencies? Did we reserve bits for something that is no longer relevant to this application use case? Release it, right? So with that, I want to point to all the fun improvements in FAPR.
36:18
I'm actually gonna do one more piece on a combined software and hardware aspect actually is actually was anyone in the session right before this in this room, the direct flash session with Peter Factor and Justin I say hand or two. OK, so this is only possible via co-engineering of software and hardware. There's a really good video on Facebook, um, on YouTube actually with Meta and Pure where we're
36:40
talking about. This and how actually anyone who's trying to do larger flash modules you need about a byte of RAM for about 1000 bytes of flash and actually if you map that out you get to say a 100 terabyte module you'd actually need 100GB of RAM. That's not what we're doing here and it actually goes to this idea of this thing called an indirection unit and you're how you're
36:58
actually tracking what the flash is actually with how the data is getting written. And because of how we actually map up here, we can do indirection units that are 10 to 20 to 50 times smaller or larger than the competition while still having the same performance. So this is hardware innovation that's only enabled by co-engineering the software whether it's Flash array or Flash Blade, Flash ray, etc.
37:24
Without, without overprovisioning, these are zerop drives again, I'm a drive guy, so I'm gonna put down the, uh, the magical click here here and just do the old school. So I'm also like click, click click. So if you haven't looked back at Flash ray for a couple of years potentially, we have been on a tear from a software standpoint with heavy continuing investment.
37:45
Now fascinating piece to me. If I embrace, uh, I don't know, some other products that start with power everything in their line or I look at at ones that are blue, etc. we talk about the engineering that we're doing here. It goes across every single flash array model regardless of the price and performance profile,
38:03
right? So in some cases when I talk about file. That could make sense for Flash ray Excel, like you see like work down here for SAP HANA, that feels like it makes a lot of sense. But some of the work, for instance around user shares, we'd love for you to put that on an Excel, but let's be real, you're probably more likely gonna put it on a flash array C,
38:22
for instance, if you will. So some of the major advances in the last year or two would be around, uh, secure multi-tenancy for files. This is around having Being able to support multiple active directory domains. I remember back to a Hay conference a while back where someone was saying, you know, you're, you're only one acquisition away from having a multi-cloud strategy or for
38:41
anyone out here who only has one domain in their environment, one active directory forest, you might only be one acquisition away from having to deal with multiple of them, right? This is just the reality that we live in. So this is about being enterprise ready. I'm always on SMB, so this is about not having disruption to SMB operations through upgrades
38:59
or fail over and upgrade is a planned fail over back and forth, right, kind of thing. User and group quotas being able to take that further. We have directory quotas, but you want to sign at a user and group level. This is a concept that's been around for a while. We're not gonna beat it to death.
39:12
And then the piece here that does matter sometimes is it doesn't matter what we do. It's more about the ISV or the application vendor, right? So video surveillance is a huge and growing market. I was actually talking with uh another product manager Tago about thinking about video surveillance even for Evergreen one potentially, but Genintech is a large player in the space.
39:33
Veronus goes into insights and visibility. I mean you. May have even been seen comprise out on the show floor part with them, but specific integrations with Veronus. And then last but not least, is there, is there anything more enterprisey than SAP or HANA? I don't think so and actually supporting that here from a file standpoint because that's what
39:53
that is what's sometimes preferred for them or depending on your environment. Let's expand the lens a little bit. So that's file on flash array, significant investment, and actually, I also forgot one piece here. We're not just doing this for fun. It's hard, it's hard work from an engineering standpoint, and we're not doing it to just go
40:11
and say, let's work down a checklist of all the file features that are out there. There's actually this intentional evaluation of what has been done over the last 1015, 20 years, what is worth recreating, maybe doing it differently, sometimes more policy driven. And some features we may not support, we may choose not to support because it didn't,
40:31
doesn't make sense today versus 20 years ago. I think if something like NDMP, for instance, there's a reason that that's not in there. There's better ways to do data protection at scale than that. I didn't get any dagger glares from the room when I said that, so, or maybe I got some NTMP, PTSD.
40:47
Expanding the lens a little bit, what's new from a larger purity on flash array standpoint? Especially if you've chosen to consolidate or have a lot of flash ray XLs, you may be thinking about secure multi-tenancy. How do I carve this up into separate administrative domains? Or maybe I'm an MSP and I want to carve it up different ways for my customers so we can do
41:08
show back for individual workloads. Um this also there can be a. aspect here so we can actually do sub delegation within the array so there's not as full access from an admin standpoint bullet number 2. And then of course we could do some pieces around performance tiering. Um, this could go into show back, charge back, scare back,
41:26
whichever one you like. I've used different ones over the years. What you think about space, ops and bandwidth. Next, replication encryption. This one to me is a little bit of a fun one. Should you ever be replicating your most critical data over a non-trusted network?
41:45
Would I ever recommend that to any of you? No. So we, we won't need to have security at the network layer. So at one level, replication encryption is for a regulatory purpose. Like it says there are regulatory requirements. We're talking to auditors and they say this is needed. Now at the same time, I'll acknowledge that
42:03
there can be scenarios where you need this for technical and security tax surface reasons kind of thing. Either way, it's available, but I actually like to embrace that sometimes you're looking at this like, mm, yes, there's a regulatory aspect, there's a technical need and requirement aspect. Last but not least here.
42:19
Who here has heard some of the announcements recently around the walls? Somebody, anybody, a lot. Let's start with Mark and Walter. There's some out there for a little bit of a, you know, poor one out here. So we are, we are doing what I think is fundamentally the right thing. We have been, as,
42:36
as you may know, a design partner with VMware around various aspects. One of them active cluster for VAs or VAs in a synchronous replication scenario, that kind of thing. We had the engineering work done on this. When the announcement came out, we are still choosing to release this for our customers that are using and leveraging VAs. If I'm really cynical,
42:58
I could say like, yeah, we should support that because of course there's still gonna be, you know, time to upgrade and there will be some support calls and questions and set up etc. kind of thing, but we're still choosing to release this because we think it's the right thing for customers and we are continuing to partner heavily with VMware from an engineering integration aspect.
43:16
So there is more coming despite what you're hearing about VAs as it relates to the VMFS and engineering advances. So One second, I'm almost in a wrap up and I got 1 2nd, so I'll do a question the answer. That's all right. Please make sure to attend some other sessions here.
43:32
Um, there's, we can dive deeper into the databases you saw another slide. There's deeper on file. We gave a, you know, kind of a cliff notes high level version here. And then of course looking at high stakes business performance pieces here. And as you've heard, please make sure to sign up for the career community.
