Interview with Vaughn Stewart - CTO36

Machine-generated from the episode audio and not hand-corrected, so names and technical terms may be imperfect. The audio is authoritative.

Hi, you're listening to episode 36 of the CTO Advisor Chats. I have to apologize in advance. We went a little bit long on this episode, but I think it was well worth it. The guest, Von Stewart of Pure Storage, talked about everything from Pure's product line and high-level offering, but more of a historical view of the industry from virtualization to cloud trends and how to help organizations move from legacy three-tier operational models to more of a DevOps slash cloud environment. I think it was most definitely a worthwhile lesson.

I hope you enjoy it as much as I enjoyed having Von on as a guest. So Von, can you go ahead and introduce yourself? Absolutely, Keith. Thanks for having me on. My name is Von Stewart. For those in your podcast listeners too who maybe aren't familiar with my voice, I'm the vice president and enterprise architect at Pure Storage. That means I'm in the SE leadership or the field's technical leadership, helping our largest customers as they transition from their traditional storage architectures and helping them understand what we can do in conjunction with their business processes, applications, or initiatives.

Prior to that, I spent 13 plus years at NetApp. The last seven, I helped lead the technical strategy for VMware and then eventually their cloud direction. I led a team at that period of time of all-stars who brought the first form of data deduplication in the market for VMware. We promoted the first use of NFS into that marketplace and actually released the first storage vendor plugin for vCenter at the time. And then some of us on the team got awarded a patent for that work, which was pretty cool.

A lot of those guys on those teams now work for a number of other storage vendors and I'm proud to say that they're helping drive great companies like Nutanix and Nimble forward. But I and some of those team members are now at Pure Storage and we're having a lot of fun over here. Well, I'm really, it's an honor to have you on because your storage roots run deep. And I think the really interesting part of it is that, especially from your days in NetApp, the innovations that NetApp and VMware did with NFS and bringing this really, I wanna call it nebulous thought of shared storage down kind of to that SMB mid-market tier where before you'd have to have a completely separate network and fiber channel to support a concept like shared storage and something that's needed to do vMotion.

NetApp, I think was really at the center of innovation when it came to a lot of great VMware vSphere technologies. And it's great to pick your brain about what's going on today because I think you have a very more so than other storage providers or other storage vendors. Your view is very VMware centric. Is that fair to say? I think that's fair to say. What's really interesting, sometimes you hear in IT that there's no net new idea. When my team drove a lot of our work 10 years ago at NetApp, that's the origin timeframe for NFS and data reduction and the vCenter plugins and array-based snapshot and snapshot backups and all this kind of great stuff that we did back then.

I'll engage with customers today who are still using those technologies and some of them are very vehement that they won't ever switch from those technologies. And the conversation that I try to have for them today is that back in 2006 through 2010, it wasn't about NFS versus VMFS. It was about scale and simplicity and ability to have integrations that you just couldn't have with VMware's clustered file system at that point in time. The way that simplicity and scale, and I mean operational simplicity and scale as well as financial scaling, cost, those were the drivers that those technologies helped move forward at that time.

Those same drivers exist today. No organization has gotten to a point where they say we're getting enough output from our employees, we have no more need for them to work faster or be more agile or efficient. That's a constant. And so those same drivers today have moved beyond the whole NFS element, if you will, for example. I wrote a post back in 2008 or nine, I believe, a blog post that said data reduction would be the norm for all storage arrays in the future.

And it is, and it's here today. The last book I wrote with Mike Schlesinger, who's over at Nutanix now, we wrote a book called Virtualization Changes Everything, and it probably came out about 12 months too late, but I had a lot going on in my personal life at that time that kind of delayed the book coming out. But our principles in that book still ring true today, which is you've got to get the stack simplified without putting the business at risk. You've got to get more out of your employees by taking the mundane and the complex out of the equation.

And when you look at, and I'm going to apologize in advance here, and you look at what we're doing at Pure Storage, particularly with VMware, right? Our driver here is to say, we can take the price point out of storage tiers that you've purchased in the past. But there was no performance need for tiers of storage. It was a financial need. But it led to operational overhead complexity when things would have a performance issue in storage. You tend to have a pretty bad day.

So at Pure, what we've done through robust data reduction is flatline the price points between 7,200 RPM drives and 15,000 RPM drives. We give it to you in Flash, and now you use technologies that VMware brings along, for example, to do quality of service and performance categorization to find its software. And you put it on Flash and you go, my bronze customers get a bronze service level at a bronze price point, and my platinum customers receive a platinum service level. The best part is it's at a bronze price point.

And all of these work on one platform that requires no tuning or tweaking in tiers and all that jazz. So it's a radically different day in storage today. And I would actually advocate for customers who are looking at any of the new storage technologies to really understand, forget what the construct is. Forget if it's NFS or iSCSI or fiber channel. Forget if it's scale up or scale out. Forget if it's an array or if it's hyper-converged. Boil it down to how does it impact your job.

And if the vendors aren't showing you a way to eliminate the tuning, if the vendors aren't showing you a way that changes the pricing model from what you've had in the past, then you're looking at the wrong technology. Even if it's a technology that you've loved because you know it for 10 years, it's the wrong technology going forward. So that abstraction is really hard for most of us that's been in the industry for 10 plus years. As we've gotten introduced to, just as the fiber channel guys have a really hard concept getting away from the challenges of iSCSI when iSCSI first got introduced.

And we have these built-in concepts or constructs of what a resilient infrastructure looks like. If we want this resilient infrastructure that has this mass array of options, then it has to be complex. You can't have a lot of options without complexity has been the mantle of enterprise IT for a long time. I think that's being challenged by the AWSs of the world. AWS's infrastructure itself may be complicated. What they provide the end users is an extraction that isn't very complicated.

So how do we bridge the gap practically from that mindset of the technology, the bits, the nuts and bolts of it is complicated. And we need to trust that to the experts, to the vendors, to the peers, to the new tonics of the world. We need to trust that layer to that group and we need to abstract and think about the services provided. How do we make that bridge practically? Okay, you've touched on two discrete points here. And this is fantastic because as you and I know prior to the recording of this week, we said we were gonna take this kind of completely off the cuff.

And this is two great topics you brought up. I think if I were to rephrase what you're asking and we can spend some time on both of these. One is around operational complexity and the complexity and the technologies you put in place. I think the other side of the equation is around what I'll call the consumerization of IT. It's that on-demand consumption interface that your end users, your organizations, the team that you support as a person who delivers IT services, it's that model that they want.

So let's break this in. Let's start with complications within the storage or infrastructure stack. You're right, there is an inherent amount of complexity in the infrastructure layer. What we are seeing and witnessing at a number of levels is a simplification of that complexity. And a lot of it, to be frank, is driven by just more horsepower, right? More horsepower in the CPU allows us to virtualize and that gave us great advancements. This mass increase in IO or storage performance or reduction of latency, you could, that's provided by solid state storage, so flash arrays, is this massive gain that we just haven't had in the storage industry for 15 years.

And it allows you to fuel a lot of innovation. Now, with that said, you still need to pull back the covers of the onion a little bit. So for example, if you, let me, how do I state this? There are vendors like Pure Storage that have focused on getting rid of the tuning parameters and adding a lot of intelligence, both in terms of the way that the storage array operates as well as how our support mechanisms work. And I'll give you a couple examples here so we can do a comparison and contrast.

So traditionally in a storage array, you have a fixed block size. And that's a model that maps down to the way that the data is written onto the persistent tier. It's some arrays has a direct correlation to the amount of memory that that storage device may require relative to the physical capacity of data it could store. It also tends to have relationships in terms of, let's say the number of your rate of what data reduction you could achieve or the number of cloning copies that you can make.

These are design considerations. But it also impacts your performance because though a storage array has a fixed block size, it has no understanding or more importantly, no influence on the size of the block that comes out of the server and the application that runs on top of it. And so in Pure, we decided to put in a lot of what we'll call intelligence or adaptive behaviors within the array. So we've got a variable length block size, anywhere from 512 bytes, which is the smallest size of a SCSI packet up to 32K.

And so when we receive an IO request from an application, if it's 4K, it sits on a 4K block on the back end processes 4K, it's really efficient. That works well on a fixed 4K block storage array, which there's a lot of them in the industry. NetApp is 4K, EMC, say, ExtremeIO is 8K, and HP3PAR is 16K, for example. Those are all fixed block sizes. With those block sizes, it means a 4K IO on Pure is pretty optimal. It's going to be pretty optimal on NetApp, it's 4K.

It's going to be suboptimal on ExtremeIO. You're going to waste a little bit of space. You're going to waste a little bit more space on a larger block like a 3PAR. But it also means the converse, which is when we receive a larger block, say, 32K, right, that's still one IO process for us. It's now eight for a 4K system. It's four for a 32K system, and it's two for a 16K system. And so that adaptive block size really helps the array scale.

And it's not a linear scale. You don't directly scale from 4K to 32K and say it's eight times the workload and thus one-eighth the performance. It's not linear in that regards, but it does allow the customer to utilize the harbor resources in a more efficient manner and take the tuning parameters out of it. You know, there are storage vendors that will let you set block sizes per volume. But then again, you're guessing at, you know, what's the construct I need for this dataset relative to this storage pool.

And as we know with virtualization or cloud platforms, you know, we publish VMs, they get consumed. We have no idea what gets installed inside of them, what the rewrite behaviors are, what the block sizes are. And block sizes, even if you think you know them, can be variable, right? So I hear this all the time as an example. I know Oracle, it's an 8K block size by default. Well, yeah, that's the size of the blocks in the database. That's not the IO size or the average IO size of all the IO that gets communicated up and down the stack because, you know, Oracle writes for every user that's logged in, writes its own log entries, and those are 512 bytes, so they're very small.

By default, all query sizes are one megabyte. And so when you start to mix multiple streams in the IO path for the application, you don't have anything that looks anywhere near that of an 8K IO stream or 8K block size. So it's a driving adaptability in the product so that you can make things simple. And what we're, if I take a step from that, we're starting to see this notion through lots of vendor initiatives of what I call a very composable and adaptive ecosystem.

So for example, you know, look at somebody like Brocade within their new Gen 6 platforms, and I think this might even be in some of the Gen 5, but they've got this notion now of like this auto zoning, and that's not their official name, but you know, this notion of like peer zoning where every host is looked at as a peer, storage devices are principal, and as you connect the, build the fabric, it'll automatically build rules and policies. So such as like when I have a VMware client provision a new data store from the storage, and that's all done through a VMware interface, the storage rate communicates the fabric, and it goes ahead and auto zones all the connectivity to the hosts, you know, the vSphere or ESXi host.

I mean, it's some really cool constructs going on, you know, underneath the covers. We've got a feature now called predictive analytics in our support that the coolest use case that I've heard of was a customer put some new hosts on the fabric, and their HBAs log into the fabric, and then they zone them to the same, you know, zones as a peer device, and we pick up and identify that there's new hosts within our zone, and predictive analytics opens a support case for the customer and says, hey, we noticed these new HBAs, keep the vendor out of this conversation, if you will, but we noticed these new HBAs are on the fabric.

We detect that they have a firmware that has a known data corruption issue, and we would recommend you upgrade this firmware to this other version or higher. This all happens before the customer ever provisioned storage or put data on the platform. I mean, it's a really powerful notion of how advanced the infrastructure technology is getting. Now, on the other side, right, on the whole, they said there's two components here. On the other side, on this whole consumerization piece, if you go and poll why money development teams have gone to AWS, it was all around agility.

Yes, people thought it was cost, but most customers are starting to understand that AWS is only cost-effective if you have ephemeral uses. Any form of persistence will cost you much more in AWS than it does in your own data center, and I'm talking total costs, not just storage costs. And that's even with iterative usage versus persistent usage. If you've got an instance that has to persist and has a data set that persists, you'll pay more. But anyways, coming back around, so Amazon really created this whole consumerization of IT element, right?

This whole, hit a portal, there's a service catalog, check what you want, put in how I'm going to charge you back with your credit card, and boom, voila, you have compute resources. I have been waiting for that in the enterprise for the last five or six years, and as much as I love OpenStack, OpenStack is not an enterprise technology. OpenStack is a telco service provider-grade platform. And to be frank, VMware, who owns the cloud technologies and private cloud technology leader, you could say, has really kind of missed the boat up until late.

I am ecstatic with what VMware's brought to market with vRealize, the vRealize suite, and you look at the things that the run books, both for managing infrastructure as well as providing a service catalog for end users, it's just off the charts. We're doing great things at Pure. I don't know if any of your readers follow Cody Osterman. com, has got great demos and examples out there, as well as now what we're starting to see come right around the corner from Microsoft with the Azure Stack, right?

An on-premise private cloud management interface that directly interfaces with their Azure public cloud offering, and so a single interface to give an end user on-demand type of experience regardless of where the resources sit. So that last mile of making something for the users has gone a long way. So again, we kind of touched base on two topics there, kind of making the infrastructure, the IT technology, smarter, behave better, scale in a more dynamic manner, to really bring it all full circle and have that front-end end user experience.

It's exciting times. So I think one of the things that also, that a lot of my listeners are looking for is the driving of efficiency. So when we look at vRealize, we look at Azure Stack, and the front-end capability that both these platforms give to the end user. I like to call it the API to the data center, basically this Amazon-like experience that we've simplified and we reduced the amount of friction between end users being developers, consuming data center resources, and then those data center resources being provisioned.

I think a lot of challenge comes through. We look at the vendor set, or I'm sorry, the analyst set, and there's this concept of bimodal IT, there's DevOps, there's all these buzzwords. And I look at it, 95% of my data center, or 90% of my data center is this legacy exchange, ERP, Oracle Stack, that I need to carve out a piece of it for this new world. I'm not personally sold on bimodal IT. The DevOps concept, I think, has some chops to it.

What's the connection or the bridge in getting to an environment in which I have this huge investment in VMware, vSphere, Hyper-V, take my management platform for granted. I have this huge investment that's driving business value today. How do I start to bridge over from that current environment to this more agile environment in which I'm supporting that new 10% of application workload that needs that agility backed by some of this stuff from Pure, Nutanix, and the likes? Sure. This is really, really topical.

One of my submissions for VMworld 2016 is titled something like private cloud costs less and is more capable than public cloud. And we pulled together an analyst from IDC, a customer. We have ourselves and we're still trying to fill a fourth slot. So maybe I'll hit you on the back and maybe you can join this conversation. But what we've done is we're going to share a bunch of data as well as technical knowledge. So it's kind of like half economics, half IT implementation level technology that demonstrates and shows multiple use cases of moving to the public cloud with persistent applications is probably the worst use of that technology from a cost and management perspective.

So to translate that for you, everything you have running in VMworld today, you have more capabilities out of the box and it's costing you less than if you translate, transition that workload into the cloud. For anything that's ephemeral, anything that's got a burst, anything that's got a spin up, spin down, then you have some opportunities in the cloud to use that at a more cost effective manner. And so we want to go through that. I would say looking at your audience, who I have to assume because of our backgrounds and the technology circles we run in are very VMware centric.

I think if you look up and you say, I've got to drive, support a new set of initiatives. I mean, I've got team members who want to develop in containers, Docker, Kubernetes, right? I've got people who want to write to Amazon technologies and launch in the cloud. I think the best thing you can do right now is to figure out how can you implement a technology so that you can decrease the level of investment and effort that you have in your current VMware footprint or Microsoft Hyper-V or whatever it may be.

Because it's not going to go away. I mean, the reality is the bulk of it. You don't retire applications and we don't retire data sets. We tend to pick up more annually and the rate at which we adopt more continues to increase year over year. So you have to look at what's your investment curve operationally between investigating new technology, getting it stood up, right? Going through whatever refinement of that technology is and then starting to de-invest that level of effort as you move it into a maintenance type of mode.

And so for all you folks who are in the VMware space, you should be in a maintenance mode for your core infrastructure. And if you're going to invest more into it, it should be on the front end, end user consumerization interface like a VRealize or an Azure Stack, right? So you want to stop supporting the backend and keep moving the value forward is what I'm saying. But you still have to watch out technologies. And so I want to pivot here and pause on this thread one moment.

Not all technologies that are new are simple and not all of them apply to all the customers in use cases as easily. We've been surprised as we've worked with some customers who have looked at hyper-converged platforms and they go, well, this must be simpler because the VMware admin can run their own storage. And we agree, VMware admins want control of storage resources. They're on the hook for delivering cloud that's highly available and performant. They should own storage devices. This is kind of like what I'll call the Exadata effect now happening for virtualization.

Hyper-converged is attractive to these folks. The top of rack storage, where they're building pod architectures that has two rack units of switching at the top and three rack units of storage, supported by like 36 blades or 36 1U servers, is pretty, pretty predominant. But what we've even seen with some of these HCI vendors that do an all-flash configuration for the performance and so they can define service levels and software. And then they add data reduction to bring down the cost, produce results of like 10 milliseconds, 15 milliseconds of latency.

And it's like, wow, you took all flash devices and you make them behave like disk arrays. And it's been really crazy. So you have to, I think, sometimes push a little harder in understanding what the capabilities of the technology are. And one of the things that I would say is for as much as we are empowering the convergence of storage and server teams, the server teams needed to listen to their storage guys, at least around minimum requirements around like data protection levels and things of that nature, because taking the minimum levels of data protection will result in data loss at some point in time.

And if you don't believe me, go talk to the old scarred, gray-haired storage guys and they'll tell you what a nightmare that is. I've had many a conversation with Howard Marks over that very topic. You know, I'm not a storage guy by default and I look at some of these software-only solutions and I'm like, you know what? This is all seeming too good to be true. What am I missing? I think you hit on a few of those topics. Yeah, yeah.

So in the flash phase, right? So first off, all flash and spinning disk has had unrecoverable error rates. The thing with flash is those error rates increase over the life of the device. And so, you know, heads up, any VMware or Hyper-V admin out there, if you're doing an all-flash config and you think mirroring equals one or false to tolerate equal one or replication factor equal one is great because it gives you great performance, you are really putting your company at risk and protecting their data.

And trust me, it's not worth the cost savings to not do a replication factor equals two or false to tolerate equals two type of methodology. Your executive team will thank you for it by never losing data. Pivoting off of my soapbox here, I apologize. Where were we at? Oh, so in agility, right? You see the storage market, for example, having all this kind of disruption occurring in it right now. So you look at IDCs, Worldwide External Disk Tracker, for example, and the top six vendors have remained the top six vendors for the last, what, five, six, 10 years, NetApp, EMC, HP, HDS, Dell.

All of them are losing market share and all of them have an all-flash array product. And I think the market's trying to figure out where's this market share going? And people assume it's going into the cloud, so we must have to go to the cloud. And actually, I would propose that cloud is healthy, well, and growing. And cloud's best use case is next-gen platforms, ones that don't rely on the infrastructure for resiliency. But most of those are applications that are web front-end facing, internet of things, collection points, right?

They have mass scaling because they need to support millions, tens of millions, or hundreds of millions of connections. They're still not the core IT infrastructure components. What we're seeing in the data center with the decrease in the, and shift in share in the storage space is that, one, when I buy Flash, because data reduction is the norm, I'm probably getting anywhere from three to 10 times the usable capacity of what I have with a disk drive. And what I mean by that specifically is I take a disk drive and I put RAID on it, and then I put write sizing, and then I put some RAID, some overhead, I'm striping for IOPS and not capacity.

The average disk drive gets about 35% utilization. An average Flash drive gets 300 to 500% utilization. So it's a radical difference in terms of the raw capacity a vendor has to sell. So I think that's, when you're looking at revenue, that's shifting the revenue, particularly at high-end systems and driving it down. I think more importantly, all of those disk systems that are retrofitted for Flash have not removed any of the complexity that's inherent in those systems. And that's why you see the new startups with very simpler models really grabbing share.

Pure, in the last quarter, Tracker finally got broken out of the group Other, which is the largest growing group on the worldwide external disk tracker. We're just above Fujitsu and just below Oracle, and the only vendor with triple-digit growth percentage. And we keep doing what we're doing over the course of the next four to eight quarters. We should get real close to that top six list. So we've touched on a bunch of topics that could probably get us talking for another hour.

A couple of things that, just in closing, that I think I should have you on in the future that we should talk about, which is the opportunity here that vendors such as Microsoft and VMware have in helping organizations with partnerships with storage vendors, server vendors, transition to this new model of IT. I don't think, and I've gotten picked on about this, about this prediction, or about the even statement, that folks like VMware are well-positioned with vRealize to really help the customers, the core, my core audience, transition to where you stated, where we get into that 90% maintenance mode and 10% in growing of that front-end application that needs the elasticity or the frictionless interface of a vRealize or a Azure Stack.

I think that's a whole podcast on itself, and I'm really surprised and really grateful for you coming in, kind of putting a little bit of credence to that concept, and that the most exciting parts of, I think, this core data center audience is that vendors, whether they're from the storage, compute, or other traditional data center vendors, are not ignoring this space. It's an important part, and it's a tough problem, and there's some really interesting solutions out on the market. Yeah, you know, it's tough for customers as well as technologists, or at least I think it's tough for customers and technologists, based on the conversations I have, for them to figure out, what do I do next?

And if you look at the VMware audience, probably most customers who are VMware customers have virtualized 90% of their infrastructure. You've got to look up and say, what's next? Maintaining the VMware environment is not where to continue to invest, and I think that's what I was talking about, like, update the technology and let it go run automagically. Getting to vRealize, getting to delivery of IT services, I would probably suggest for most of you, if you run a VMware shop, that unless you're at a really large scale, where you've got some financial costs that you really need to address, OpenStack might not be for you, right?

OpenStack is, it iterates at a much different pace, it's much more community-driven, you've got to really invest a lot of time to be on top of things with OpenStack. Though there are some great vendors that are simplifying it, you know, vendors like Mirantis and Red Hat. Mirantis has got it super simple. But, you know, if I was a VMware admin today, I would throw all, I know this is self-assertive, I'd throw all flash in the back and make it dumb and simple.

I'd focus on two things. I'd focus on vRealize automation, and I'd focus on go figuring out a development platform that I would want to run my company on, be it Docker or Kubernetes or Mesosphere, those kind of seem like the three that come up most in conversation, and go advance my skillset and have a platform that's ready so that when my executive team says, we're going to go to the cloud, and you go, great, are we going there for containers in the development team?

I've got a platform here ready, I've got my cost model, let's go compare it. If they say, no, we're going to move some of the VMware or Microsoft to the cloud, you go, great. Here's my environment, it's highly optimized, it's got low touch, and here's my cost model. I think if you're a technologist today and you're not factoring cost, you're probably doing yourself a disservice. And even if you go like, no, really, I'm really a diehard technical guy, it also means you're really smart and you're probably not doing your team and your staff and basically the fruits of your labor, the justice of representing them financially because your leaders above you aren't looking at everything on a financial basis.

All right, well, I really appreciate you taking out the time and I know we had a date on the calendar a couple of times and we finally got to it, but I think the conversation was well worth it. It was extremely enlightening. Where can people find you on the web and Twitter? Oh, hey, yeah, great. So first and foremost, thank you for having me on this, Keith, it's been way too long. We run into each other at the conferences, but we should find time to have more conversations.

And obviously this one's tech-centric, but not deep. Maybe the next one we can go deep if you prefer. com. It's the virtual storage guy, which is also a domain you can hit. Or on the Twitters at vstewed, V-S-T-E-W-E-D is where I'm at on Twitter. Looking forward to catching up with you in a couple months as we get into trade show season. Looking forward to it, thanks a lot. Thanks, man.