00:03
All right. So welcome to this session. Um We uh it, it's obviously described in the catalog as the the pure storage road map in, in particular around flash Ara. So we chose to give it a headline of speed of innovation. And the reason for that is that it's not only about talking about the content of our road map,
00:22
but it's actually also about how pure continues to innovate and bring that innovation to market. And so as we uh we get started with this session here, my name is Peter Scoop and I'm responsible for flash in particular PG. So the software features and with me is partner in Crime Bill. So I'm the general manager of the platform business unit.
00:46
So I oversee the uh hardware systems design and I've been with pure about 10 years. So I've seen uh a lot over the years. So we're gonna do a little history lesson and then uh talk about the current products and then talk about the future products. All right. So it, it's not a coincidence that you see hard software together here. Um Our hard software road map is carefully
01:10
orchestrated. Um so that uh a single entity being the flash ara allows us to, to continue to bring innovation to our customers. And so I wanted to start out with just a very brief um statement here and I think this is an important one to look at is is very much about innovation. And if you look at the quote here, the most powerful force in business is the pace of
01:36
innovation. When you bring innovation, you bring value, increasing value to the customers to you. That's what this is about. And there's a reason we often call evergreen subscription to innovation because it's about our systems. Always getting better, has invested significantly in bringing innovation to the market faster and and doing it in a much
02:01
more agile fashion. We structured our engineering and development processes as well as the organization around continuous innovation. And so we actually at the point where we bring new features to flash array on a monthly basis, I don't know how many of you have noticed that, but this is how pure maintains the growth and innovation leadership in the market space.
02:23
However, innovation only matters if you really bring it to market. And so one of the things I wanted to do here was just briefly um show you what that means and what we've done since last year. If you accelerate and fast forward to today, this is what has been delivered in flashing. So quite a few features here, some of these are incremental features,
02:45
some of them fundamentally changes or add new use cases. And so examples around this would be unified blocks and files, NVME over TCP active workload, auto safe mode. And we could go on common for these features is all that they are available as a software upgrade for free. So all part of the ever green program, so you might be sitting and thinking about how do you
03:12
actually get to adopt? I mean, if you get a new release every month, could you actually do that? Could you actually get all of those features in your system? And someone else would probably say no. And so that's why we come out with a new release model. So starting with 64, we have um two different trains of releases.
03:32
There's a continuous innovation train which is called or it used to be called our standard releases. This is where on a perpetual basis, we come out with monthly releases, new features and we'll just continue to do that and you can stay on that train forever. You can also choose to be on our long uh life releases which provides a level of stability.
03:54
So a long life release doesn't get any new features, but it lives for three years, fully supported. And so you can choose between those two releases. And more importantly, you can change your mind later on, you can switch between the stability, the long life releases and the future releases whenever
04:11
you would need to. And so this was introduced with 64, you'll see the first long live release with 65. And so this is part of how we set up the tools and what we do to enable this innovation and really be able to utilize it. But it's not all about the software.
04:29
It's also about um the hardware and how pure comes together with a unique flash based uh architecture and how we do the tight integration. And so let's see about the hardware platform from Bill. All right, I thought it might be fun to look back in time and then think about what decisions we made about how we would design systems in the future and then how that brought
04:55
us up to today. So we're in the way back machine here. OK. This is the road map that we used to show in June 2013. And it's, it's special to me because I joined in July of 2013. So this is I walked in, this was the road map that we were showing.
05:12
And so what you would see is that we had our fa 300 moving to the 400 we had our usable capacity was growing. The usable capacity included ded Dulic location data reduction. So 5 to 1. So in the second half of the year, we were going to move to a 35 terabyte system and we would have 5 to 1 data reduction we'd have over 175 terabytes system and everything was great.
05:40
It's up to the right things are improving and we're moving very quickly with scale and so on. But that's not what my team saw. We saw a dead end, it looked like a dead end to us. We were using off the shelf components and they weren't ready to scale and they didn't meet all the requirements. We had an amazing purity operating system,
06:03
the same operating system we use today. But this is how we thought about the hardware. This is what it felt like to us. We were using servers off the shelf. We were using AJ BOD that we were procuring from a third party and we were using off the shelf S sds. Who knows what computer that is?
06:20
That's a Commodore 64 people who are younger in the group probably don't realize that keyboard is a full computer. It's a bit computer. I don't mean to kick it around too much. They sold 17 million copies of the Commodore 64 and it was for, you know, a few months, it was kicking Apple's ass, but that didn't last very long.
06:43
And we saw the same problem in our uh in our systems. We weren't doing all the design work enabling all of the value of pure with the hardware. So we decided to take a step back and think about what is important to us. OK? What are the basic design principles that were important to us. So Evergreen, Evergreen is the promise that the
07:09
system will always get newer. You can ndu non disruptively and move to newer and newer hardware, your data stays in place. It's really, at that time, it was really a revolutionary thought. Most customers didn't think we could pull it off. And you heard in the keynote today about the, the customer who had seen 85 non disruptive
07:30
upgrades, but the hardware really wasn't ready to do that. We had non disruptive upgrade upgrade in 2013, but you know, you had to do some gymnastics and it, there was a lot of cable moving and yes, it worked, but it wasn't very easy and there were some risks involved performance. We knew if we were using off the shelf components, our performance was slow.
07:55
We knew if we did hardware and software co design, we could go faster and then simplicity also back to the cable example I gave you before opex is the real cost, right? How hard is it to use the equipment? And the underlying system architecture wasn't enabling that. So we said, ok, these are the principles that are really important to us and we made some
08:17
decisions going forward, ok? And these decisions really set the table for the for the products you see today and they were big decisions at the time and some of them were really kind of risky. So first off we do fully optimized systems. So if you think about it, it's really easy for a company to go get a server, put a new bezel on it, write some software and send it out into the market place.
08:40
It's another decision to decide we're going to do this from scratch. And every component that's not needed in the system is out and every component that needs to be in the system is optimized. So we have access to the best um all of the, the best components and the best architecture and the best block diagram. So we made that decision, we decided to take full advantage of flash.
09:01
Now in 2013, making the decision to not use off the shelf S SDS was really hard, it was really hard, the SSD industry was taking off. There's many um companies offering to supply great S SDS, but we saw that had more potential than was an SSD and SSD was built to look like a hard drive and it had to serve many, many customers. We had a specific use case.
09:29
We knew if we took over the SSD design that we could unlock the full potential of, of NAN flash. Now, this was a big deal because we had on my team, we had to develop a firmware development team, an SSD development team, a characterization development team which are very high tech teams, expensive teams. And it it wasn't obvious to us that we could
09:50
pull it off. But we have and then we had to do modular systems. So one of the reasons you're seeing so many products come out so quickly from pure storage is because we built some basic building blocks and we're using those building blocks in different configurations to offer all of these products. That's why something like a flash blade E and a
10:13
flash A eight E can come out in the same year, right? Compared to the competition, the pace is just different. And that was really where we were focused. Can we get this out and then can we iterate and really drive pace forward? So we're just going faster. All right, so fast forward. 10 years.
10:34
Now, we have the direct flash module. We shipped our first direct flash module in um 2017. And when we did it, we were nervous, we had gone this new path to do our own SSD. And so we dipped our toe, we did one of our products X and we had the DFM and then we realized we had a clear advantage and we started to broaden it and widen it.
10:58
And now we've been shipping for six years now. The control of neon, the data placement onto the NAN is, is controlled centrally and globally. So there's one the operating system is working across all of the jobs at once and making all the decisions across the entire population within a different, within a system. There's no compromise QLCQLC is really,
11:22
really hard to manage. OK. It, the NAN cell is, is um very high scale, very complicated. It has a lot of quirks and every QLC offering from the NAN vendor is different and we have to adjust to that. We are able to do that with our architecture and we offer systems. You'll see that have full QLC so many of our competitors, they'll throw in some faster
11:46
drives, some right cash. No, we have systems where every drive is QC, there's no hidden flash S SDS. Today, they take a portion of the flash, they hide it, they use it to compensate for background operations that are compromising their performance. We expose all of the flash.
12:06
So our 75 terabyte module, if you were to buy that off the shelf would be like a 62 terabyte module, distributed NV ram. This is not a right buffer on the drive. It's a non volatile ram location. The OS can write to the NV ram on any DFM and then it can use the data to send it to some other DFM. It's just a dedicated location for the OS to make decisions about how to do
12:34
how to have non volatile rights and do right, commits back to the, to the host. And then because of all these innovations, we can do very high density. So all these things added together and the fact that the DFM is single use only in our systems allows us to build bigger and bigger drives.
12:58
So this is a snapshot of the product line today. We um if you've gone to some of the other talks, the keynote you heard about these new products. Again, if, if you look at this, the L came out in 21 the two R four products are coming out right now and s was last year. That's a really fast pace for completely new product line. The DFM technology works across all of those
13:26
products. And now we're announcing E so we've seen tremendous pick up on Flash Blade E which was announced earlier this year. And then now we're releasing flash array E. So from 1 to 3.6 petabytes down for flash array E and then from four petabytes up on flash Blade E and they use the same QLC module.
13:49
And when you see us doing larger and larger DF MS, this is the product line where they're going to emerge first because this is the capacity, um hard drive take out product line. All right. Awesome. Imagine if we had told you this a year ago that this is where we would be from both the hardware and the software perspective.
14:10
It's pretty exciting and we can't really wait to get the flash A into your hands. So, um by now, some of you may be thinking about, OK. So the headline of this session was about road maps. So, well, let's stop talking road maps. So innovation is core to pure and um we're always looking for new ways to do
14:32
things. Um We're willing to try, we're willing to potentially fail, right? Just to explore. And so some of the things we do won't always work out. That's OK. And so when I was trying to pull this session together and, and look at the road map that we're gonna discuss here,
14:49
I was trying to think about new ways of actually doing the road map. And so, um you know, maybe the portion of uh of the product management role where we do road maps is obsolete. So let's try and ask uh chat GP T. So what does a new world map look like?
15:06
Imagine what it came up with? So here's what the, what it turned into. So first of all came up and said, well, in order to maintain a cutting edge and innovation here, there's a set of things we should do. So let's just try and run through them really quick.
15:22
I know that this may be too small text for you guys to read at the back of the room. But uh nonetheless, uh this is what I got back. So innovation in flash technology. Well, pure is built for flash from the ground up. You just heard Bill. We've been building DF MS for more than six years. So sorry, G GP GP T.
15:42
Uh Not really a prediction here. Not really a road map. Item, second part scalability and performance, well largest and most dense drives in the industry. Latest generation of platform almost across the board all for performance. Again, a bit ahead of GP T. So you know, maybe a little bit more learning to be done here,
16:04
data management analytics. Well, we're very focused on simplicity. Um Strong Dr we introduced the direct compression module for offload. We do pure analytics in forms of assessment recommendations. We do workload simulations. We just introduced anomaly detection. So I would give ourselves a couple of checkmarks here as you can see going through.
16:29
Next one is uh cloud integration. How many of you have played with CBS CBS for AWS Azure fusion cloud operating model, cloud snap as a way to expand out. And with all the latest introductions we've done there, strong cloud focus. And so I'm going to give us a checkmark on that as well. Next one security and data protection,
16:54
we've been very focused around safe mode, auto unsafe mode. As I said, we're just introducing anomaly detection. We introduced a ransomware SL A. If you haven't seen the pure one discussion, please go go to that session, you'll find it interesting. So we're definitely checking the marks on that part of it.
17:14
And then finally, customer centric approach. Well, the pure culture is built on this. We pride ourselves of listening very carefully to you. Each employee is directly incented on being customer and and centric. We have the the highest validated NPS score in the business. So the conclusion that you come to in terms of
17:37
this experiment is sorry, GP T go back and learn a little bit more um because this isn't really a road map. So that leaves us with basically going back to the old model of doing things if this will work for me. And that's called the good old George Crystal Ball.
17:59
So P MS are not entirely obsolete yet, but it's worth having some fun, right? And um and innovation is about trying things and potentially failing, getting feedback and that same approach in our road map. And so you'll see both short and long term things on the road map and I picked a few areas if you think about how much was on the road map from last year and what we delivered,
18:23
we're not gonna be able to talk about all of it here nonetheless. Um picked a few items. All right. Not so much. OK. Flasher rate files. How many of you tried Flasher raid files? How many of you have you heard about Flasher raid files with our latest launch?
18:43
Oh, we got work to do not all of you. All right. Well, so flash files um is all about the truly unified approach that we introduced and we're trying to bring innovation to the file segment of the market. We're not the first in this market, but we are coming from behind and um with very fast pace of bringing innovation
19:06
into that. We're not creating new file protocols, but we are bringing a new way of managing your data. We are bringing a new way of thinking about it. We're questioning the legacy and saying there's a better way we can do better than what you're used to. So we're bringing simplicity, we bring automation, we bring visibility to your data
19:28
that doesn't exist in the market today. So some of the components we introduced with alarms is global storage s being able to dynamically make decisions. Don't be locked into any particular design that you set up for. We're doing workload lar management, we're doing automation. We try to assume everything we possibly can about your workloads and do the work for you.
19:51
We do VM at WE M management. We introduced manage directory and integration directly into the V plugins. We will continue to innovate, we will continue to expand. We have a lot of things coming this summer and a lot of things that are targeted for completion of the protocol stacks. And so by the summer,
20:09
you will see that we are mostly complete with both SMB and NFS. You'll also see full security implementation, scalability, et cetera. So lots of things on our road map. And so our plan is to bring farmers to every flash array, your choice, whether you use it, but we want it in every possible customer environment. And so though we are coming up from behind.
20:33
We're coming in fast and our a our ambition is very clear here. It's nothing short of. We want to be number one in files. We are going to displace the legacy. We're gonna build a solution that makes you want to go to flash files. So that's a component of a road map.
20:54
Next part of it, active workload, how many have heard about active workload? We introduced it in 64, right? So this is all about data mobility. We're obviously doing a play on the naming here. The active component for pure represents the non disruptive nature of the future. So this is about non disruptive mobility, even a cross platform. So you can go between AC and an X as an example
21:20
and move your workloads nonn disruptively from a host perspective. And so we also introduced multiple sync connections. So if you're used to using any of our sync tools, you can now have connected to up to five different arrays and the whole point is building a mesh of flash arrays that allows you to do mobility in any fashion you would want.
21:42
After that, if you're familiar with active cluster, we're going to make mobility of the volumes, we're going to allow you to a volume, move in and out and between parts non disruptively. You do not have to unstressed data to a synchronous setup relationship and sync it up and let it just come in completely automated non disruptively.
22:06
So beyond that, what we're really looking at is completely ad hoc data mobility and allowing you to manage that. We also want to make sure that you can operate with um a policy based approach. And this is where fusion come into place where your automation of data placement and the mobility of it becomes a function of that cloud operating model,
22:29
be able to move workloads around and preferably let the system do it by itself. Other things we're looking at self service, how many of you have tried self service upgrade? Just a couple. You should really take a look at self service upgrade, we call it SSU internally. And so this is all about allowing you to do unscheduled upgrades whenever you want,
22:57
you simply go to pure one and you upgrade your platform to the target release you want. It's as simple as that it does everything that you would normally see done by our support team. It's been completely automated into the system. So really a strong experience with this, the behavior that we see is that customers are taking releases 2 to 3 times as fast as what they previously did.
23:24
And so this is about bringing the innovation, not just to market but bringing it into your arrays and making that happen. So self service is not just about clicking the button and letting it happen, it's actually about being able to help with decisions as well. And so provide the information.
23:43
So we're just introducing policy driven upgrades. So you can start seeing where this is going. We're trying to automate the upgrade of your fleet. And so with the policy driven upgrade, we help you think about how you want to think you want to utilize the different releases we provide and help you make the decision and then help you implement the upgrade of those
24:05
platforms. Long term, what this turns into is a couple of things if you're used to your windows or your Mac upgrade where it just happens during the night and you no longer care, that's where we want to go. And we think we're gonna get there some of you is gonna sit and think no way I'm ever letting the system do that by itself, right?
24:24
It's OK. We already move borders well beyond what any anybody anticipated and we'll keep moving them. The part that's going to come in here is that the information available and how you consume information is going to change. It's all about having all the information available to you to complete the task as self service.
24:45
And so with that, you're gonna start seeing pure one evolving into providing a lot more information. So think about this right? Today we read a PDF or release notes. Why, why does that make sense anymore? It seems like past century kind of technology, we need to be able to do context sensitive information to you.
25:03
We need to do it in the picture of your fleet, your systems and we need to be able to generate the information you need. So we think about it at information as a service that comes as a part of the self service approach. And so we really see a lot of evolution in this area of the system.
25:21
So if you haven't tried it, this is the next thing to go do on your array. All right, next part security obviously front and center for most people. And so part of the security has a lot to do with ransomware and safe mode. And so with 64, we introduced the auto on safe mode. And so we also introduced the data protection assessment.
25:47
And so data process assessment in pure one gives you a full fleet picture of where you are protected and where, where you may not be. And so next step is that we are going to introduce things like anomaly detection. So you see the first inkling of that in pure one, it's going to grow much faster and better than what it is today. But a good start.
26:10
The other thing is we're going to turn on safe mode on all your arrays by default. So when you upgrade to the latest 6465 release, you're going to get safe mode unless you make the decision deliberately to say I do not need protection against ransomware. And so we're forcing you to make a decision because this is so important. But we're going to bring a lot of tools with it. So this isn't guess work.
26:37
We're going to do things like self service pin management. We're going to update and improve the eradication time. We're expanding the use cases where safe mode works really well, including integration with backup systems. All of these things are coming, but it's all about deliberate choices.
26:53
That's what the component is about here. And so you'll see us evolve. We've done a lot of other things in addition to safe mode. But this is the one that's usually the top uh talk uh topic here. Um Other things we done like UB key for AD right?
27:11
So integration with modern ecosystem components all coming available, we will continue to integrate into the security ecosystem. We will continue to make enhancement for your ransomware protection. That means simpler, faster recovery, it means much stronger integration for the purpose of detection and it continues to build out the solution that we are offering to you with it.
27:39
Next step, scalability. You heard Bill talk about what he's gonna do with drives, right? And I don't think it's a surprise even that he's gonna claim 300 terabyte drives, who knows, maybe even more in the future. So what that means is that so far compared to the introduction of flash array
27:58
purity has scaled about 100 X. And what that means is all the configure objects have scaled that much. And so as we continue to scale, right, with the current four petabyte on series, we um we need to continue to increase the objects. So, examples of what we're doing and what we are just about to start talking about is we
28:19
increased the number of snapshots with 12 X on the large platforms. And so we're really bringing the ability to consume the storage capacity with the optics and how you're operating your workloads to the platform all through a simple software upgrade. So you kind of see the pattern here. And obviously, as we go to 300 terabyte drives and you can do the math,
28:43
we start talking about 10, 12 petabyte worth of of raw storage. You're gonna have to change how you do data management. You're not going to be able to maintain the same paradigm with that kind of scale because at that point, we scaled 1000 eggs compared to where we started out. So a couple of things comes to mind there. It's a testament to the flexibility of purity
29:05
and what we do with it. But it also means the way you think about storage is changing. So with that in mind, you know, I talked a lot about the software here bill, maybe we should look at the road map for the hardware, the platform side of things. So this is our latest design.
29:27
We're just, we're just coming out with this in 2023. OK? The, uh this guy has been around for a long time. II I looked up the uh history on hard drives and hard drive was, was uh first developed by IBM in 1953 1953. It's 70 years old this year.
29:48
So you have to really think about that. OK. Because the transistor was invented in 1947 and this was invented in 1953 the integrated circuit didn't come until 1958. OK. Five years after this was invented 70 years and look, it's gotten to a point where it's hard to scale. It's a mechanical device. It's very high energy,
30:15
uses a lot of power to do what it does and it's prone to failure and it also creates a lot of waste when you have to replace one. So seven years has been an awesome run, but I think it's time is over. All right. So here's a chart devoid of numbers. Um But it's gonna show trends and I'm going to talk about why it matters so much to
30:39
us. OK. So the hard disk drive is still getting larger very, very slowly. It's really lost its ability to scale. And if you watch the hard disk drive predictions from the industry, they're always predicting a rise in scale and the actual rise in scale is about half of what they predict if you look over the last 10 years.
31:03
So it's very, very slow and it's slower than the companies that make hard drives predict then there's S SDS sds are getting larger. They don't have all the ability to scale DMS because they have all these different use cases and they have all these limitations with a flash translation layer on drive. The drive is doing all these background operations. It's very hard for them to scale.
31:26
But also there's another thing here, SSD vendors aren't motivated to scale. They don't have a lot of customers for 100 terabyte drive, right? It's a very big investment. But at pure storage, every time we make a larger drive, half over half of our bits move to the larger drive almost immediately.
31:45
Today, most of our bits are shipping on the 48 terabyte drive where our competitors don't even have drives that big. So you see that we're highly motivated to make larger and larger drives, but the market isn't driving the SSD vendors to make huge drives. So we have a clear advantage here.
32:03
Now, the reason I show this is this is the motivation behind all the hardware innovation that's coming up, we are going to scale the systems. When the drive gets larger and larger, you have to scale performance because some applications performance per terabyte is the key thing as the systems get larger and larger, we have to focus on that.
32:23
You'll see us do things like offload engines. So we announce the direct compress accelerator DC A which is a compression offload but there are other offloads that we'll be bringing to market in the future. Also, we will have ways to end U between product lines. So already you can end U from an X 90 from the X line to the XL line.
32:47
And we'll start to expand on that so that you can move between product lines and move media between products and so on. The same drive that goes in E of flash blade goes in flash array, the same drive that goes in flash blade s goes into flash ac. So this is the main motivation and the driver between the hardware innovation. We have really blown out the product line in the past two years,
33:15
we used to be X and flash blade. And now we're XL X ce two models of flash blade, the S 500 the S 200 flash blade E. So we have a full portfolio of products. Now, now we're going to scale up the performance over time and meet the challenges of these larger and larger DFM. OK. So I have numbers on this slide.
33:36
All right. So I want to give you a real world example of what's possible with future drives. All right. So we're shipping the 48 today, we announced the 75 we're working on the 1 50 in the lab right now. So 100 petabyte example, 100 petabyte example for flash blade,
33:54
you have to buy multiple chassis, right? And so in the first example, which is what we're shipping before the 75 comes out is we have, it would take 75 chassis 13 racks, 100 and 40 kilowatts to run this 100 petabyte example. And then with 75 you see it drops to 90 kilowatts. Think about a terabyte per watt in that system.
34:18
Ok? And off the shelf, one terabyte drive takes about 12 watts today. Just one drive just to put it into perspective. 1 50 we moved down to 2.43 terabytes per watt, we move up to 2.43. So it doubles obviously with the size of the drive, we moved all the way from 13 racks down to four racks with 150 terabytes.
34:39
That's an amazing change. This is not over a long period of time, this is over a short period of time. And then with 300 terabytes, we get all the way 100 petabytes in two racks in 22 kilowatts, 2 11 kilowatt racks, 100 petabytes, five terabytes per watt. This is a seven X change.
35:03
It's an enormous change and it requires us to not just make a larger drive but also move the systems forward to keep up with the pressure of these larger drives. So I wanted to give you one real world example to think about I know saving power is so important to everyone. So this is a place where we're focused all Right?
35:26
I think that is the end of the presentation. So we wanted to go to questions if um I don't think we have any microphones for questions. If you ask a question, we'll try to repeat it and so everyone can hear it. So any questions? Thank you. Thank you.
