Transcript
Mike Matchett: Hi, Mike Matchett with Small World Big Data and I'm here today talking about, of course, storage, one of our favorite topics. But we want to talk a little bit about some of the advanced kinds of storage that are on the market today. Some of the cool new things you can do. Not your traditional sands, not your traditional, even HCI solutions and software defined solutions. But you know what? What can you do at the extreme edges of the storage challenge today? So I've got neriad. Just hang on a second. We'll get right to it. Hi, Andrew. Welcome to our show. Hey, Mike, how are you? So I understand, you know, from a little bit that we talked that, you know, basically what you have is block storage. You know, that's been around forever, but its storage has been designed for sort of this extreme level of data before we get into the bits and bytes of that, though. What's your background and what drew you into the storage industry? Andrew Russell: Yeah, thank you for the question, Mike. So I've been in the technology space for about 30 years now. I got my start up in Canada designing broadcast and TV and radio stations. And really what got me into storage is when I joined Harris Corporation out of Florence, Kentucky, they were a systems integrator. Um, at that point in time, digital was only just beginning to catch on in the world of broadcast and media. And, you know, storing a lot of data in that space was super interesting to me. And the opportunity came up to join EMC. And at that point in time, EMC was developing a broadcast media server. It was a bit of a skunkworks project going on in the engineering organization. They were looking to bring it out to the marketplace, and that is where I got my start in storage was at EMC for about 15 years, doing various different jobs over there. Went on to Pure Storage, where I was there for seven years, and I've been here at nerd for the last two and a half. Mike Matchett: All right. So quite, quite an auspicious career there in storage. Most people have heard of EMC and pure I bet. So I understand I understand the theory though is kind of a kind of a down down under country, at least for us North Americans. It was founded in New Zealand, right? Andrew Russell: That is correct in 2015. Mike Matchett: Yeah. How did you manage to hook up with a New Zealand company? Andrew Russell: Um, so about two and a half years ago, I was looking for something different to do. You know, I absolutely do love the storage space. You know, when I joined pure, they were doing something very innovative and different, very disruptive to the marketplace, coming out with the first all flash array. And I came upon, knew it and saw that they were leveraging GPUs and they were using GPUs inside of storage. And that caught my interest because I had never heard of that before. It wasn't really a use case I had thought of using GPUs for. So I wanted to learn a little bit more. Mike Matchett: All right. Yeah. Yeah, we definitely want to. It's catching my attention, too. Like GPUs in storage. Not for the graphics, I'm assuming, but for some other efficient purpose. Let's just talk about the use case before we get in. Before we get into the bits and bytes though, this the storage then that you're aiming for. When said extreme before, really you were talking about some sort of large scale and some sort of large volume. What what kind of edge of the market? What are some of the use cases for that kind of storage? Andrew Russell: Yeah. So we are focused really on what I would say are four core markets. One is that of high performance computing. So think of some of those activities where you're leveraging parallel file systems like IBM Spectrum Scale or Gpfs or Luster or Beegfs object based storage. So very fast, you know, sort of very as it sounds like performant type type of workloads. Think of anything in the media space. So we've got content ingest. You know, cameras are now up to eight K. That is very large file format, you know, takes quite a bit of performance to actually work with those formats. So think, ingest, think post-production rendering, think video surveillance or streaming type applications. Other use cases would be in the world of backup and restore. So these are typically very large sequential streams of data coming in. And when you need to restore you need to restore those fairly quickly. And then the world of active archive. So think medical imaging or any kind of again areas where you know it's not a level of tape, it's not a super, super active data, but it's all that data that sits in between. Mike Matchett: All right. Pretty wide variety of of of markets there. But everyone can relate to backup and recovery given the ransomware and everything we have going on the threats out there. So this is something I think most people will be interested in learning more about. So let's go back to that, that idea that that struck you first, which is the idea of using GPUs in storage when when you started putting GPUs in storage, what to what use are they putting it to? How does that fold into IO? How does how does all that floating point acceleration accelerate and enhance storage. Andrew Russell: So we're actually using the GPUs in a couple of ways. So in traditional block based storage systems from a protection scheme they pretty much all use a form of raid. And Raid was invented back in I want to say 1986. I believe we're actually leveraging a ratio coding at the block level inside the storage system. And this is something very different. And we're actually using the GPUs to calculate that ratio. Coding the GPU we're using is a mid range Nvidia GPU. It's got roughly 10,000 cores. But that is a lot of processing power to be able to calculate those ratio coding. So that is one way that we're using the GPUs. The other way is that we have separate read and write paths in, in and out of the array itself. So we're using the GPUs on the writes or on the side. So we're able to very rapidly move the data in through the GPUs and write it down to the media below. And then we use the CPUs on the read side of things. And what this allows us to do is essentially treat both the reads and the writes very, very much the same. So we're not worried about read write ratios because everything is not going through the CPUs. We have separate paths, so we're able to handle the ingest as well as we handle the reads. Mike Matchett: That's interesting. So you can separate that out and focus on optimizing write and ingest. So I take it with the GPUs and the erasure coding. Then what we're calculating is all the math that goes around creating the the N plus one or N plus 20 or whatever the number is, erasure coding bits and farming them out properly in a scenario like that. Let me just ask, does does the does the actual end media then matter much to to performance? How you know, what do you what do you what are you using as your sort of actual storage media underneath that. Andrew Russell: Yeah. So our ultra storage system is actually software defined. So we use 100% off the shelf components. We don't have any custom components made. So in theory from a media perspective you could really put anything in there. What we've come to market with today is an all in. Speaker3: The solution. Andrew Russell: It's not a flash based solution or, you know, sort of a hybrid, but we're leveraging Https, and we're actually able to drive up to 20 gigabyte per second of sustained read and write throughput capability using Https. So we are, you know, essentially approaching flash level performance with that. Mike Matchett: All right. So that's pretty good because you know the workloads you're talking about high volume, high capacity large files, right. You know if you if you think of putting a flash system against some of that, they get the speed. But the capacities don't match up with the economics in a lot of cases, I would guess. And so here you're able to say we can give you that flash light performance using erasure coding across lots of lots of hard drives on that which which is a pretty interesting thing. And and guess one of the things that you know, I was interested in is when I, when I look at that kind of thing, the ratio coding you get a lot of fast rebuild then. Right? I mean if you if you have any sort of failures, the whole environment can participate in, in rebuilding that. Andrew Russell: Yeah. Essentially what we're doing and we're not going to get too technical on this is we're doing a wide stripe either across 102 or across 204 drives. So when a drive does fail, it needs to be rebuilt. We've got all of those other drives that are actively participating in rebuilding the data. Mike Matchett: Okay, okay. And you can get some other goodness with with the erasure coding too. What's sort of the largest system then than you would would, would see people using it for today. Andrew Russell: So today, the largest system that we ship as an appliance is 3.6PB in capacity. Mike Matchett: Well that's pretty that's pretty big. Guess that would that would store a lot of my video anyway. All right. So we could probably dive into some of the technologies here all day long. But you've got something to talk about. You've got something to announce here. And that is providing ultra io the solution here more as a service rather than as a piece of investment, a commodity sort of investment here. How does that work as a service. Andrew Russell: So essentially we would deploy a storage system to the customer's location, either on premise at their facility or in an MSP or colo of their choosing. The customer essentially would manage the system as if they own it near it retains the title to the asset, and ultimately the customer is subscribing, if you will, a subscription to actually use, you know, certain amount of capacity on the system itself. Mike Matchett: All right. And we've seen a couple other companies doing this saying, you know, NetApp and pure Dell with Greenlake and so on, where they will, will will run this as a cloud service. How much, how much do you how much of a challenge is it for you then to pre-position equipment? And is is that something you guys are prepared to optimize for, you know, for both your clients benefit and your benefit? Andrew Russell: Yeah, absolutely. We certainly see that the subscription models, it provides customers flexibility and choice. You know, traditionally if you purchase an asset or just treat it as a capital asset acquisition, you're carrying on the books, you got depreciation, you've got all those implement implications. Most as a service offerings are typically treated as opex. However, you know, that's obviously subject to the customer's internal accounting policies and rules. But it's really about providing customers flexibility and choice in terms of how they actually acquire storage. Mike Matchett: All right. Because you know, we we know if we look at storage as a service and even just looking at something that most people might be familiar with, which is Amazon S3. Now, when I go to S3, there's like 7 or 8 different variants of that to choose from and more coming all the time. If I'm not really well versed in storage, I might have a hard time picking the right thing there or settling on the right allocation. And I'm betting a lot of other folks who use that don't. So is, you know, I know you offer a lot of options, but is it easy for someone to navigate or make changes to, or do you kind of have to sort of predefine when you buy it? This is what I want this capacity for in in tranches. Andrew Russell: Yeah, certainly. As you mentioned, there is a, you know, look at as a level of complexity in terms of some of the options that are out there. Oftentimes you may be asked to look at your workloads. You know, you may be asked, well do you need block storage or file or is it object. And you may be subject to different reserve capacities for those different types. We've actually tried to simplify that complexity. The storage system itself is able to handle block, file and object all at the same time simultaneously on the system. And as such, we can actually take some of the complexity out of our as a service offering. By working with the customer customers, we we we do resell. I should point point out 100% via the channel. So we work with our channel partners and the and the customers to understand what it is they're doing, what are the workloads that they're going to be putting on the system, and then work with them to come up with an optimal capacity amount as a starting point? We're not worried, you know, if they want to choose, if they need so much of a block and so much for file and so much for object, we can handle all of that on the same array. Mike Matchett: All right. So so that makes a job easier for you. But also then gives a customer a lot of agility as they as they get this into their environment, they don't have to be worried about getting it exactly right. Specified from day one they can say I need, you know, we have a larger growth picture and we can grow into that any way we we need to which I like that. Talk to me a little bit about then the complexities of licensing this and paying for it in a lot of other arrangements, you know people. Look for way complex models of how storage should be paid for over time, even in the even in that storage as a service kind of perspective. How are you guys approaching that? Andrew Russell: Uh, actually fairly straightforward. Um, essentially, the customer just has to select the, the services they want, right. Block file object. And then it is we provide a single price per gigabyte or a price per terabyte per month based on the capacity that they need. That is the same price per terabyte for the reserve that they're subscribing to. And even for the space that they go into, once they go into the on demand or into that space, we actually charge this the same amount as the reserve, and it's based on what they actually use. Um, and we would build the on demand either a quarterly or monthly in arrears. Mike Matchett: Okay. So that that is that if someone really wants to think of this as an OpEx model, is there anything is there anything that's out of whack or does this really look like that full cloud opex. Andrew Russell: Uh. Again, I'm not a I'm not a financial accountant. And obviously we would leave that up to, you know, our customers financial, um, you know, policies and rules as to how they would deem that. But essentially, since we retain the title, the customer is subscribing to a use a certain capacity over a certain term. And, you know, at the end of the term they can simply return the system to it, or they can extend the term. Most would deem that as an opex type transaction. Mike Matchett: Right. So really getting to that utility consumption model of but pre-positioned somewhat according to what they, what they've ordered. Uh, yeah. So that's interesting. So you can be delivering that sort of utility consumption model and helping customers get over the idea of, of pre-positioning what they need ahead of time and what they use over time to be pretty optimal, I would think, and really help them focus on things. Would you have any sort of best practice recommendations for someone looking at putting storage into their environment? What what should they be considering and and what would lead them to look at myriad? Guess. Andrew Russell: Um, think think that one. You want to look at how you're using it, right? It's not one size fits all. There are a lot of storage options on the marketplace. You know, there are all there are hybrid. There are certainly all flash arrays. I think it's really looking at understanding the workloads that you have and optimizing the right storage solution for those workloads, because not all storage systems will necessarily work for workloads. Um, so one, I would just really, you know, take a look at that second, um, you know, really understand your growth patterns over time, you know, sort of understand where you are today, what does the business intend to do, and how is that going to grow and change over time? Because what we don't want to do is get yourself into a situation where you know you're going to grow very, very rapidly and you now have to deploy multiple systems, right? That you now have to manage to handle that. But at the same time, you don't want to, you know, purchase a petabyte of capacity. If you really only need 50TB today and you're only growing at, say, a couple of terabytes per month, now, you really, you know, sort of overinvesting in terms of what you need. So it's important to sort of balance right sizing the environment, but also looking to the future in terms of the growth patterns of the business. Mike Matchett: Right. And I think you're talking about a couple of things there that I would put under that sort of complexity. Simplicity. Factor where you want to make sure that you're not proliferating storage services. Because even if you have storage as a service, as an OpEx, your staff multiplies times seven for seven different kinds of storage is going to be expensive as well, so you're not really saving on the back end if you have to do that. Okay. This is pretty, pretty fun. I'm, you know, would love to dive more into how more specifically and technically, GPUs accelerate some of those functionalities you were talking about. But I get it. In general, you've got a erasure coding which is functionally numerically deep, and you've got some other things going on on write ingestion to put things together. And so basically accelerating that sounds like sounds like the right approach if someone wants to learn more, Andrew, about this solution, the product ultra ultra IO or ultra IO as a service, what would you point them out? Andrew Russell: So certainly they can come to our website which is nerdio. So there's a vast amount of information available there. We welcome anybody that wants to evaluate or POC or trial the storage solution itself. We are very open to deploying a system at a customer's environment for them to test that, test it out and actually see how it works in their environment. Mike Matchett: All right, all right. So got that offer folks. Thank you for explaining that to us today Andrew. We hope to have you back. Andrew Russell: Thank you very much, Mike. Mike Matchett: Check it out if you need some storage. And particularly in those use cases, we talked about where you have large capacity utilization for media, for large files, for backup, and for a couple of the other things that we talked about. Check out near you. It might be just the thing you're looking for. So take care.
