Grey Beards on Systems - 69: GreyBeards talk HCI with Lee Caswell, VP Products, Storage & Availability, VMware

Episode Date: August 17, 2018

Sponsored by: For this episode we preview VMworld by talking with Lee Caswell (@LeeCaswell), Vice President of Product, Storage and Availability, VMware. This is the third time Lee’s been on our show, the previous one was back in August of last year. Lee’s been at VMware for a couple of years now and, among other things, is … Continue reading "69: GreyBeards talk HCI with Lee Caswell, VP Products, Storage & Availability, VMware"

Hey everybody, Ray Lucchese here with Howard Marks here. Welcome to another sponsored episode of the GrayBirds on Storage podcast, a show where we get GrayBird bloggers together with storage and system vendors to discuss upcoming products, technologies, and trends affecting the data center today. This great-bred on storage podcast is brought to you today by VMware and was recorded on August 17th, 2018. We have with us here today, Lee Caswell, VP Products of Storage and Availability at VMware. And this is the third time we've had Lee on our show. Why don't you tell us a little bit about what's new at VMware, Lee? Yeah, thanks so much, Ray and Howard. Always good to talk with you guys.

And, well, we have a lot going on. And, you know, for many of you are getting ready for VMworld coming up just two weeks away. It's the big party. It is. And within that, you know, HCI and vSAN powered products all the way from vSAN ready nodes into VxRail, VxRack, what we're doing in VMC. I mean, what you're going to find, right, is that the expanded definition of HCI is now a core part of virtually every VMware initiative. I mean, it's just super exciting to see how this new storage architectural control plane is now extending itself into the full software defined data center. So what do you mean by an expanded definition? Extended, expanded.

Well, I think, you know, when we first started talking about this point product of, hey, we could go and put storage control, basically a storage controller, into the hypervisor, right? And now storage and compute, right, because of fast CPUs, low latency internet, you know, our network connections. And all those drive slots servers come with if you use them or not. Well, and you've got Flash, right? So now what you could see is you had a scale-out design, right? That was kind of the flexibility of virtual machine control to now adding in storage. So that was first, right? But I think of that more like, that's kind of like AM radio, right? I mean, it was a small start, right, from a sound standpoint. Then what's happened, happened right is now we've expanded that so that makes some of the other guys bing crosby oh come on maybe well so you went from let's say am radio to fm radio right with uh expanding the full on-premises deployment with you know a full

stack approach and that full stack approach man it was not just storage and compute, but we're adding virtual networking in, fully wrapped by alerts, monitoring, management, automation. And that basically then said, we've got a full software-defined data center stack, if you will, that could run either as software on standard servers or as an appliance model. And then when you really expand the lens,

you know, they open the aperture even farther, you know, the IMAX view of HCI says, well, wait a minute, like that same software stack, now that it's a software stack, could be running on any location. So it could be running in the public cloud, could be extending out into the edge or above the stack,

I'm extending from typical, let's say, applications, enterprise applications running in VMs to, well, now I've got container-based cloud native apps. And so it's this idea that the full stack is the most important decision you'll make as a company. And it's incidental where it's running and what's running on it. STDC at the edge? Why not? I mean. In a robo solution or something like that?

I always thought the edge was a great place for HCI. Yeah. Yeah. Well, I mean, look at the edge. I mean, you don't have storage administrators sitting out at the edge. You're lucky you have IT people sitting out at the edge. Not alone a cabinet. You have the plant manager who comes by once a week and changes a tape. Or if you have somebody centrally now looking and saying, hey, can I do that from a SaaS automation standpoint?

Can I look and start monitoring and managing from a central location across all these remote sites? That then becomes a fully automated, right? Basically, I call that live infrastructure, right? Yep, and we'll replicate all the data instead of making the guy change a tape and now we can run it all from home. Yeah. So you start thinking about this software-defined means

at the edge and, you know, you're going to have to be separated from the edge. I was listening to Pat Gelsinger talk today about the three rules about why the edge is going to matter. So one was it was the law of economics. Well, he started with laws of physics, laws of physics, laws of economics, laws of the land. Those are the three. So physics, you know, right? Speed of light.

186,000 miles per second. It's not just a good idea. It's the law. There we go. Right. So that's one reason. Second one was economics, right? You're going to be wanting to process information locally, right? In a distributed manner and not bring everything into a, you know, a fully rented environment. And then third was just, you know, laws of the land mean that especially data people, you know this, right? I mean, data, you know, once I start trying to move data, it doesn't move very fast. It's not free to move. And there's sovereignty issues, right, to be considering. So all of those reasons mean that the edge, I think, is going to be a huge opportunity with, what is it, 50 billion endpoints by 2020. That's just my home, right?

I mean, I turned out, I don't want to talk about it, but I turned on like smart thermostats and smart fire alarms. I was terrible. I got more endpoints in here than I ever conceived of. But before I go down that path, you also mentioned beyond the data center or beyond. Where does containers fit into all this HCI stuff? Yeah, here's what we see, right? We're actually in the process of really saying there is no private cloud anymore.

I mean, there's a hybrid cloud, which is the idea that you're going to want to tap into the massive investment and features and flexibility that are available in the public cloud. So we see this interconnection across the hybrid cloud as being the de facto model going forward. And so the question is, how do you engage with the public cloud? I mean, one train of thought is, oh, well, that's just going to open up a mass exodus from on-premises into the cloud. A lot of data people know that some people tried that to start with and found out that it was very flexible, but it was also very expensive. There's the mass exodus to the cloud, and then there's the bill that comes, and there's a mass exodus back from the cloud. Repatriation right so so then maybe what it is is you know there's real value in containers

that's different i'll suggest than what happened with open stack oh very much right i mean how would you how would you characterize the difference well open stack was a way to run applications containers are a way to build applications. That's correct. And so there's so many things like video transcoding where I don't want to keep a VM running all the time. I just want something ephemeral to do the work and drop it where it's supposed to. Yeah. Yeah. And our, our view of this is that, you know, there's an infrastructure level, which is where VMware has always provided resiliency, simplicity of management, flexibility, right? That, that environment is one where, you know, while it

sounded self-serving at the start, like, Hey, containers running in VMs. I mean, that's in practice what we're seeing because you want to have the flexibility of the development value of containers, but still have a common developer-ready experience, right, that goes all the way from legacy applications into cloud-native applications. Well, and to some extent, you don't want to have the developers start on $150,000 worth of hardware so they can run on four bare metal servers. Right.

Precisely. And, you know, to the modern developer, you know, the infrastructure is almost incidental. I heard a funny story the other day about talking to a cloud native developer who was running on Cassandra and said, well, what storage are you using? I said, well, Cassandra. He didn't know. No, no. What's the storage? And he's like, what storage are you using? He said, well, Cassandra. He didn't know. You're like, no, no. What's a storage?

And he's like, what are you talking about? So the idea was like the infrastructure, right, is completely abstracted. Well, is that really all that different from the Oracle developer who's writing code to run against Oracle, who doesn't know what kind of storage the DBA is providing for him. Yeah, but the DBA knows, right? I mean, somebody there is doing RMAN and all that other stuff, so he knows what sort of storage is there. And if you set up Cassandra, somebody did that. But I think

my point is the database engine is arguably part of infrastructure. Yeah, I think that's right. And so what HCI does that a lot of people haven't quite realized is it reunites or basically by rehydrating the storage back into the server platform effectively, right? You're basically bringing the management of storage attributes closer to the app itself. And so you've got a common management, which, you know, when you separated those, I mean, that was in the early days of virtualization and put separate external storage.

I mean, that was great to go and enable the motion, DRS, HA, and all that. But, right, the idea that when you can bring it back in, you know, you're basically reuniting the application and the infrastructure together. So you've got a faster way to go and have the infrastructure respond to changing application. So question for you guys, right? Actually, you know, so as you're thinking about HCI and we talk about this like expanded view, I mean, how do you think customers are looking at hyperconvergent? You know, starting out, it was kind of outside the data center, but we're now looking at mission-critical applications. I mean, big install.

First of all, I think that the HCI industry has reached a level of maturity that it had to before. Oh, God, the appliances that are out there nowadays are pretty serious stuff. Not only that, the whole SDD stack being available in multiple places in multiple instances i mean it's just it's uh it's sort of you've built out this ecosystem over time that's that's you know i hesitate to say but it's it's it's taking over the world i i was talking to one customer just at the flash memory summit who just wants to build out an infinite number of HCI clusters. It's here's, you know, our building block is a rack that's 16 nodes of HCI,

and we're just going to lather, rinse, repeat as far as we need to. So there, I mean, it's gone from an edge case to a tool in the toolbox. And, of course, there are a lot of architects who pull the first tool out of the toolbox and go, look, a hammer. All my problems are nails. And so it's the only tool they use, which, you know, is how things go. It simplifies their deployment, simplifies their worldview, simplifies updating, upgrading. I mean, it's just a it's just an easier way to manage.

Yeah, but our friend has another simpler way to manage. Really? Yeah, VVols. I've got a pure flash array M10. It took me 20 minutes to get VVols up and running on it. Once VVols is up and running on it, I run my VMs through SPBM. Storage policy-based management for those, right? Yeah. And it's so interesting. Yeah. Which is the same way I do it in vSAN, where I say,

I want this VM to have this level of protection. And we're looking at vVols and a lot of people would say, wow, vVols. I remember hearing about that. And what happened? Well, we've just continued to add more enterprise-ready features for vVols. A year and a half ago, we added support for array-based replication. We've got, you know, we'll be talking about adding SRM support into vVols. I think SRM is the last major piece before I can say, why are you running VMFS on your array? You should really be running vVols. Why do you think that? What are you thinking?

How are people thinking about vVols? Or how should they be thinking about vVols? Well, unfortunately, too many people aren't thinking about vVols at all. Yeah, we talked about vVols two or three years ago. It was a big thing coming out. And we thought it was going to be a major, major endeavor. And all the storage guys are going to support it right away and all this stuff. And it just didn't get adopted.

Some of the first implementations weren't the best in the world. It didn't support replication. And the part that I found surprising is, and a lot of people are still running Vsphere 5.5. We should have known that three years ago when we looked at vVols and said, this is the best thing since sliced bread, that it would take three years for people we looked at vVols and said, this is the best thing since sliced bread,

that it would take three years for people to get to it because they had to do a hypervisor upgrade and 95% likely they had to do a storage system upgrade before both ends supported it. And in the enterprise data center, that's a three to five year process to get both those upgrades done. Speaking of 5.5, 5.5 is going end of general support here soon, right? Yeah, September 16th, I believe, is the date.

So yeah, a month from today. Is there a great big countdown clock as we enter the solutions exchange at VMware? I think it should be in the VMware old countdown clock. Yeah. Well, VMware actually mailed little LCD countdown clocks to some customers. I saw one. Well, you know, one of the things that, you know, a lot of customers are looking at is, so tell me the compelling reason, right? Why I want to upgrade. So certainly a lot of value been added in 6.0 and past releases, right? But to the extent that, hey, a lot of the early functionality on the compute specific vector was implemented, they're looking saying, well, you know, there were switching costs. But now what you're looking at is this upgrade

process, right, to go from not just vSphere, but to vSphere plus vSAN, offers an opportunity to go and move, take a server refresh, and enable a storage upgrade, and particularly storage upgrades into all Flash. And Howard, you were at the FMS show, right? So we're just watching as flash prices tumble. I'm predicting about a 40% to 50% decrease in SSD prices over the next year to 18 months. NVMe SSDs?

Regardless of interface. Really? It might be even more for NVMe SSDs because they sell at a bigger premium, but that premium't have a – that premium doesn't come from parts cost. It comes from market. But now that we're coming to the end of a long glut and – actually, we're coming to the end of a long shortage. We're entering a glut. And our friend Jim Handy has explained to us several times that that means that there's going to be a precipitous fall. Exactly. And so as that comes in, what's interesting, right, is when it was expensive, then people had to think, well, how do I justify this on a performance basis, right, by application?

But now what's happening, right, is over 70% of our systems are all flash already because customers who are deploying HCI tend to be basically loading on whatever applications come on and they're not planning, right? Because nobody's good at planning anyway, right? So the idea that I'm going to plan ahead and figure out a controller for five years ahead, I mean, that's just not realistic. And so- And the best thing about all flash compared to hybrid is that it's some, the experience is so much more predictable. Well, customer sat is off the charts, right? Predictable, consistent, low latency, and you can load on pretty much whatever

you want. Right. And so you've got a really interesting way to stop managing performance. And, you know, the analogy I use for this, by the way, is, you know, when you remember when capacity was a limitation and you remember those defrag tools, we're great beers here, right? Oh yeah. I think I wrote one or two. Do you remember how much time we spent to go and defrag, compress your drives? Why? So we spent all that time because capacity was a limitation. Well, because capacity was more expensive than time. Exactly. And we have been doing the same thing on performance forever. Oh, no, but we've reached the point where performance, yeah, sometime this year,

that performance won't be more expensive than time anymore. Exactly. You know, that's it. You could write a paper on that, Howard. Yes, I, well, I could. That's a really, I mean, that is the, go ahead. I mean, the history of computing is the cost of human time compared to the cost of machine time. We punched cards because being online was too expensive. People were cheaper than running terminals.

And if you look all through the history of computing, it's all about machine time gets cheaper, human time gets more expensive, and so we make computers easy to use. Well, and this has been interesting. When you moved the storage portion of this HCI, right? It is now on a software basis. That means that we're not bounded at all by hardware transitions.

I mean, if you looked at VCN, day zero support for the CPU Skylake, day zero support forMe, DayZero support for Optane, right? What we have is we've got roughly 50,000 server developers, right? Taking the latest in CPU, media, and network technology, dropping that off at our door. And we're enabling that as fast as it comes here. And our development efforts aren't bound to specific hardware implementation right it's really anymore i mean and even if you look at uh you know new storage class memory devices right from into uh yeah the optane optane dc dims and uh netlist is doing something with flash it's a it's an exciting opportunity to think about,

hey, finally, right? The industry actually moves forward when you can parse and kind of decouple development, right? And so what we've done is we've been able to decouple the storage software and data management as well as the integrated stack now. All of that development, right, is decoupled from any particular hardware and innovation curve. Yeah, yet you get to take advantage of all of that development is decoupled from any particular hardware innovation curve.

Yep. Yet you get to take advantage of all of them. So any innovation that comes around is going to end up in… The pleasures of software defined, huh? There you go. Hey, gents, this has been great. I had pleasure to talk with you guys again. Howard, is there any last questions for Lee? No, I think we got it.

Lee, is there anything you'd like to say to our listening audience? You know, when we come back to this HCI and the stack, I think a lot of HCI, when we thought about it as like an AM radio station, it was like, all right, this is a point product, maybe an appliance, right, that's going to sit outside the data center. Now what's happening is consider the full stack and how that full stack, right. Deployed all the way from edge to core into the public cloud and across any application, right.

Gives you the full investment protection. It's the most important decision you'll make over the next five years. So look forward to seeing everyone at VMworld. And thank you so much, folks. Okay. Well, this has been great.

Thank you very much, Lee, for being on our show today. Thank you. And, and thanks for being on our show today. Thank you. And thanks to VMware for sponsoring this podcast. Next time, we will talk with another system storage technology person. Any questions you want us to ask, please let us know. If you enjoy our podcast, tell your friends about it, and please review us on iTunes as this will help us get the word out. That's it for now. Bye, Howard.

Bye, Ray. Bye, Lee. Bye, Ray. All right. Until next time. Thanks.