Grey Beards on Systems - 58: GreyBeards talk HCI with Adam Carter, Chief Architect NetApp Solidfire #NetAppHCI

Episode Date: March 16, 2018

Sponsored by: NetApp In this episode we talk with Adam Carter (@yoadamcarter), Chief Architect, NetApp Solidfire & HCI (Hyper Converged Infrastructure) solutions. Howard talked with Adam at TFD16 and I have known Adam since before the acquisition. Adam did a tour de force session on HCI architectures at TFD16 and we would encourage you to view the … Continue reading "58: GreyBeards talk HCI with Adam Carter, Chief Architect NetApp Solidfire #NetAppHCI"

. Hey everybody, Ray Lucchesi here with. Howard Marks here. Welcome to another sponsored episode of Greybeards on Storage podcast, a show where we get Greybeards Storage system bloggers to talk with storage system vendors to discuss upcoming products, technologies, and trends affecting the data center today.

This Greybeards on Storage podcast is brought to you today by NetApp. As recorded on March 5th, 2018, we're very glad to have with us here today, Adam Carter, Chief Architect for NetApp SolidFire Group. So Adam, why don't you tell us a little bit about yourself and what's new at NetApp SolidFire? Sure. Thanks, guys. Happy to be here. Just a little bit about myself. I've been part of the SolidFire team since the very beginning. Let's just say since the A round for SolidFire. So deeply involved with a lot of the product development over the years within SolidFire, my focus has been almost entirely on building out the new NetApp HCI product set, basing it off of the SolidFire technology that was acquired.

So most of what's new in my world, while we continue to mature and work on the SolidFire all-flash array, most of my focus has been on utilizing a lot of those developments to also advance the HCI technology. Hey, Howard, weren't you guys at Tech Field Day together? Yeah, Adam did a presentation at Tech Field Day where he talked about the NetApp HCI solution, but more significantly talked about HCI as an architecture. And he broke the world down into three variations. We run the storage layer in a VM, we run the storage layer in the kernel

or the base operating system, and we run the storage layer separately. The SolidFire HCI is interesting in a couple of ways. First of all, it's the only HCI solution I can think of that came out of an existing storage vendor and product. Nutanix and SimpliVity and VMware with vSAN basically started with a clean slate and said, we're going to build a storage layer into HCI. And SolidFire took a different approach and said, let's take what everybody knows and loves about SolidFire and make Hci out of it and and you know my really my only complaint is about the solid fire approach is if the h and hci stands for hyper converged it isn't that the hypervisor doesn't manage the storage process and they don't share those CPUs. But as I'm sure –

A virtualized system in there someplace? Well, there's a host running a hypervisor, but unlike a vSAN or a Nutanix, the CPU that runs the VMs isn't also running the storage process. They're more segregated. There's disadvantages to all that stuff, Adam. Yeah, absolutely. I mean, the discussion we were having the other day was a really good one. I thought that was a lot of fun to talk about the different approaches in HCI. I think what we've

done at NetApp is definitely different. We had our own take on this idea of how do you deliver simplified systems, all in one stack, a one throat to choke type HCI system. And in my opinion, the way we ended up building it has its advantages and has some differences in what you get out of the system than most typical HCI systems. Now, I also happen to think too that while we have some advantages and people can build systems differently, I've never seen a customer come up to me and say, you know, I just, I have to buy a new system where my storage processes and my apps all share cores. It's super important to my business. Nobody says that. So we didn't care about it when we were building an HCI system. They all talk

about how easy it's going to be to buy and install and manage and how fast they're going to get business, how it comes out of it. It's much more about the use of an HCI system than it is about the nuts and bolts and what layer you stack on top of what layer. So we're evangelizing for that, of course, because everybody else preceding us did things differently. It doesn't mean it's the only way to do things. There is still the, you know, the virtualization guys want to buy a solution that doesn't have any part of it with a big storage label on it so that they can not keep the storage guys out of running it. So there's some political issues. It's operational kinds of aspects of it, I guess. Yeah, that comes down to company by company, who's got the budget, who's got the decision

making power, et cetera. Who's got the decision-making power, et cetera. Who's got the juice. Right. Yeah, so to speak. And I think that differs depending on the type of customers you're selling to. And I do think there's – not all HCI customers are the same in the market. There isn't just one type. It's a popular demand. I feel like it's almost – I feel like the word revolution is used way too often, but there's just a really high bar across the whole industry now for lots of users and lots of different types of use cases and lots of different environments demanding this new level of simplicity. And I want to buy the

whole system. I want to stand it up fast. I think they're just sick of a lot of the old complexities that existed before CI or HCI systems started simplifying any of this. So these different HCI systems are going to appeal to different types of customers. And for sure I can buy the, you know – we want to eliminate the six days of the storage priesthood drawing pentagrams and lighting candles. Wait a minute. I didn't say storage priest. Yeah. We both used to, Ray.

I know. Thank God we got away from that sort of stuff. Right. So, Adam, the SolidFire HCI solution then has – it's system today is 2x4. That's what I call what you were just saying, a 2U system. Fits four blade-like systems in it. Blades is the wrong word. High density. Yeah, high density.

Thank you. Because they don't have the type of – There's no backplane. No infrastructure. Right. There's not – the chassis isn't smart. The chassis is sheet metal with power supplies and fans. That's why it's not blades. So everybody probably knows what you could picture there. Those are pretty common commodity type of systems today. a small footprint and a small entry cost without sacrificing a lot on the hardware side. They're just fitting a lot into a two-use system. I actually think they push the limits of what's

possible in density these days in data center type deployments. So we definitely utilize that type of hardware. You have compute nodes and storage nodes. And the storage nodes are a smaller form factor version of what the solid fire afa is we didn't water it down at all to put it inside of the hci system so effectively you're taking a whole solid fire sand environment and and shrunk it down into you know some number of servers in this two by two by four configuration yeah and i think that's that's that's getting to the root and shrunk it down into some number of servers in this 2x4 configuration? Yeah, and I think that's getting to the root of the biggest difference between what we built in HCI and what I think most people built in HCI. So when I look at a lot of the HCI systems out there,

and this isn't at all a negative statement, I think it's just depending on who you are and how you set out to solve the problems of the HCI market, you come at it from a different place, right? If you're a startup or if you're a hypervisor owner like VMware or Microsoft, you're going to come at it from a different angle. And what most people have done in the HCI market is taken commodity servers, taken a hypervisor, looked at that and said, okay, I need to add to this a storage layer that's good enough to make this a full infrastructure. And so they either code a new scale out storage system from scratch, or maybe they go grab one of the good open source ones as

a starting point and they customize it from there into their own storage layer. But they're essentially working from this mindset of starting from zero and working up to a sufficient storage system to run HCI. And in our case, as probably, at least to my knowledge, I think we're one of the only storage systems to then go towards HCI. We're probably not the only ones. That's just me thinking under my own lamppost. But coming from where we are, we're not going to start all over again and build a new scale-out storage system. We're like, hey, wait a minute. We have the best scale-out storage system that we've matured for the last seven years. And it has all of these capabilities that a lot of the HCI systems out there wish they had as well as we do. But the time it takes to harden that and build that is really tough. And we're looking at it going, okay, well, we can try to fit that into

a VM, or we can try to do some other things to it. But here's all the negatives with that. Or we could take it just as it is and simplify, you know, make it a little bit smaller, make it fit in the right footprint. So it's kind of like we're starting from this full fat enterprise storage system, and in simplifying it to make it an HCI storage system or starting from zero and trying to code a new storage system to be good enough for HCI. We came at it from two different sides of the spectrum. And I think you end up withrich enterprise storage system in that HCI. What about the operational side of this, Adam? I mean, so, I mean, you know, if you look at some of your competition, because it's sort of integrated into the virtualized environment and stuff like that, they seem to have some advantages from an operational

perspective. Do you provide some additional software enhancements to the standard SolidFire solution? Gosh, yeah. This wasn't just a bundling of stuff to get it out there. We, of course, we weren't starting from scratch to get management because we had to have a lot of great integration and management as just a standalone storage system. That's kind of a competitive thing as a storage system, right? How well integrated are you into VMware and how good is your plugin and all that stuff. So take that as the base we started from, and we had to take it a long way to reach the bar of, okay, here's a complete end-to-end HCI system. So that was a lot of work on how is the installation done? How is that simplified? How is that one quick wizard process that sets

everything up? That all had to be built. And we also had to improve a lot of the integrations that we had in hand to make them one seamless single UI with the full management stack inside of it. So it did take a ton of work. You can't just bolt those two together and say, great, we're done. Well, you guys did get to start with a pretty good vVolves implementation. Yeah, we did get to start with a lot of really good integrations, but I don't think we had ever done, well, what's the way for me to think about it? Before, when you thought of it as a standalone storage system, it was kind of customer by customer, like, okay, this customer is using vVols, this customer is using our QoS integration, this customer likes the VC plugin. You didn't necessarily know that everybody was going to run all of those seamlessly end to end with all the monitoring and such. So there was a few gaps there that weren't just satisfied by it. And we had to really improve the idea that all of that had to work together just perfectly in one UI and not be four different integration points. I remember talking to some of your fellow engineers

when the idea of this first came up. And my first reaction was, wait, you're going to run SolidFire in a VM? The thing I like best about SolidFire is the QoS. And if you don't own the CPU scheduler, guaranteeing performance on a per VM or per volume basis is going to get difficult. Right. So that was one of the things we loathed giving up. If you look at it, so we could have just said, okay, hey, everybody in the HCI market builds it out of a VM.

Let's just throw away all thoughts about architecture and build it out of a VM because that's what you do, right? And we didn't really do that. We were looking at it and going, hey, the second we put it in a VM, some of our favorite attributes are lost. Some of the things that really make the solidifier system as good at, and here's the thing that really matters to me, the things that make it as good at multi-tenancy

and consolidation as it always was, you lose by putting it in a VM. I can't, if I'm sharing cores and sockets, et cetera, with other applications, and if the storage stack doesn't own the sharing of that, you know, the hypervisor does. The hypervisor is the one deciding who's using what CPUs and when and how they fight amongst each other. Well, you should have just written your own hypervisor. Yeah, we could do that five years later.

So, oh, no, everybody just picks up KVM and rebrands it actually. So the thing we, we looking at that, we were like, that's, that takes the best out of our storage system. Like you're saying, Howard, what we wanted was instead we, what we had to give up is we couldn't build the lowest entry point system in the market. It's really hard to build a sub 50 or 30k entry point system without putting it inside of a VM. There are ways for us to go lower down there with the current architecture, but that was okay with us because it gave us the highest scale of system. And it also gave us the idea that we can take more than one use case. So go ahead and use our HCI to put tier one applications, multiple tier one applications, multiple tenants on it with

absolutely no worries about performance. And that's pretty unique. A lot of the HCI systems I've seen in there in the market that share cores, they do more point solutions. Here's my HCI in this office for this use case, and here's my HCI solution at this location for this other use case. I think they're trying to get better at it, but, but I think we've got a better approach to tackle that. So back circling back to something you said before, I mean, this is a fully functioning solid fire all flash array sand solution. That's been shrunk down and put into this two,

two you by four server environment, and it's got everything that the old SolidFire solution would have had? I mean, plus more, right? Is that how this all plays out? Yeah, we lost nothing in doing that. The, the, the two by four HCI version of the solid fire system is just as capable and powerful as the, the AFAs. It comes in smaller sizes. So the two actually work perfectly together too. So frankly, we, we, some of this, we just started to realize as we started to work with more customers and plot out more systems, we'd go, oh, gosh, okay, you're bigger nodes than the HCI nodes are. They can go ahead and switch to those just to get the efficiencies of the bigger nodes. They're the same system. So I can use any of the solid fire nodes as storage,

what some of the other HCI vendors would call storage only nodes? Yeah, absolutely. And something I didn't really hit on as well as I could have, it's like my head was down in the sand of working on HCI. And then we started getting more and more questions as we were working with more implementations of people pointing out, hey, wait a minute. So are there any limitations to me using that as an AFA? Or can I say hook up this other environment I have and use that as a SAN? Because it's got the storage and performance available and I could use this other use case on it.

And we just looked at it and shrugged and go, no, there's no reason. I mean, that's the full SAN sitting there. It's not locked into this HCI config. Go and hook up whatever other ESX servers you have or whatever other, I don't know, OpenStack you have, anything. Anything on the HCL. So you don't have the limitation that many of the other HCI vendors have

that in order to add compute, I have to add more storage horsepower at the same time. And this is something that's really always bothered me about most of the HCI solutions is well you know I need one more server and I could buy a server from Dell or HPE for six thousand dollars but you're going to make me buy six or eight thousand dollars worth of licenses to connect it to your hyperconverged storage or you're going to make me buy a node from you for $20,000 that adds storage to the capacity. So I like the flexibility. Yeah. There's no limitation here of the compute in this solution. They are,

not to oversimplify it, they're ESX servers. And all of the management and everything connectivity-wise is, you could put any other ESX server on this. That's no problem whatsoever. Going all in with the two-by-four architecture is probably where a lot of people will start if it's maybe a new implementation or maybe kind of greenfield. But I think most customers are actually going to have a current ESX system

made out of who knows what servers, add HCI into that seamlessly. They can use the storage with their current servers. They can migrate storage over. They can add other servers that aren't part of our HCI system. They all look like ESX servers to the management layer. So there's not a lot of religion about it has to be this HCI system with this color bezel on the front or nothing. Well, I don't know that religion is really the word I would use. I think lock-in is closer. Okay. I think the important part about being, you know, to make this really, truly an HCI solution, it's the management software that ties this all together, the support that lies behind that and that's if there's a problem with and i'll call it the netapp solidifier hci um do do i do i call netapp or i call you know vmware or how does that play out yeah you you call netapp you bought an hci system from netapp we have monitoring of the entire system compute esx XX, et cetera, all of that monitoring is wired into, yes, absolutely. So, so we, we put a lot of effort into, in my opinion, when you look at HCI and what customers

like about it and why it's so popular, one throat to choke is an important piece of it. If you sell a system as an HCI system and somehow there's three, four vendors involved in the support of it, then I think you've missed the mark on the simplicity that customers really desire. Yeah, that's one of the reasons I'm not all that big on the software-only solutions. Well, here's our software solution. Run it in a VM. It'll turn you into HCI. And when something goes wrong, how many vendors do I have to call? I always thought that was an oxymoron too for maybe slightly different reasons.

I think I get, I mean, obviously it's popular for its own reasons. Maybe it's a different customer type that I just haven't fully understood yet. But when I look at a software only, truly a software only HCI solution, second, the customer is having to think about, hmm, what hardware do I put this on? It feels like old-school infrastructure. You're having to think through CPU and RAM and how do I build it and how do I install it? Wait, I'm right back to the beginning again,

as though I'm installing ESX on this thing. And it doesn't feel like HCI to me. Well, even worse, this means that I as an end user customer have to keep the expertise of somebody who can make all of those decisions on staff. Correct. Annual call, card, all that stuff. Yeah. Okay. Well, this has been great. Adam, are there any last words you'd like to say to our listening audience? Just appreciate you tuning in and hearing us out and learning a bit more about the NetApp HCI solution and what makes us different. Great.

Howard, any final questions for Adam? No, I've got it pretty well. Okay. We would definitely like to have everybody take a look at Adam's session at Tech Field Day. It turned out to be one of the better sessions we've seen in quite a while. So I'd definitely look that up if you get a chance, and I'll certainly put a link in the podcast. Well, this has been great.

Thank you very much, Adam, for being on our show today. You're welcome. Thanks for having me. And thanks to NANF for sponsoring this podcast. Our next podcast, we'll talk to another Data Center Systems and Storage Technology guest. Any questions you want us to ask, please let us know. And if you enjoy our podcast, please tell your friends about it. And please review us on iTunes as this will help get the word out.

That's it for now. Bye, Howard. Bye, Ray. Until next time. Bye, guys. Bye. Bye, Adam.