Transcript
Mike Matchett: Hi Mike Matchett with Small World Big Data. I am here today talking about one of my favorite topics which is really hyper convergence. Its you can call it by many different names, software defined infrastructure. You can talk about it as as a way to cluster resources, virtualization hypervisors and all that stuff. Uh, but we are here talking today with Nodeweaver and we're going to look at what they've put together over the last dozen years in almost secret, uh, to create a solution that is going to blow your mind a little bit about what can be done with infrastructure that you already have lying around. So just hold on a second. Oh. Hey, uh, Carlo. Hi. Welcome to our show. Carlo Daffara, CEO & Co-Founder, Nodeweaver: Nice to meet you. Mike Matchett: And Alan. Of course. Welcome back. We've had you on before talking in various various modes. Um. Alan Conboy, Field CTO, Nodeweaver: Great to see you again, Mike. Mike Matchett: Uh, so I noticed you're both wearing the same, uh, same shirts. Obviously, you're not working together over there at Nodeweaver. Uh, and, uh, we've got the famous, uh, hyper converged infrastructure lab there behind Alan, if you're looking in the background there. Um, so let's just start a little bit with, uh, just just introduce yourselves just a little bit. Like what got you interested in working on hyper converged infrastructure? Uh, and, you know, why is this an area that you really focus on? So Carlos, tell us a little bit about you. Carlo Daffara, CEO & Co-Founder, Nodeweaver: Well, my history actually starts in the world of research. I mostly worked for, uh, research funded by the European Commission. And, uh, many years ago, we have been tasked to create a set of prototypes that would help improve the reliability of IT infrastructure in places like rural Europe, where you may have not Connectivity. You may have no IT support staff available or not. The last generation of hardware. From there, we work at several years with several partners across the world and ended up with the prototype, and we decided to spin it off and make a Nodeweaver out of the blue. Mike Matchett: Yeah, I'm just fascinated with this entire story. Once you dive into it about how you bring lessons learned from the supercomputing side of it down to the almost exact opposite end of it, which is out to the edge at, you know, like you said, rural villages where there's probably not even an I.T. guy in the village, uh, on there. So we'll get into that in a little bit. Uh, and Alan, um, you know, coming off a long history of creating, uh, converged converging infrastructure and testing out stuff like what got you into Nodeweaver. Alan Conboy, Field CTO, Nodeweaver: So Carlo and I go back many years for obvious reasons, shared interests, etc. I actually stumbled across conversations he and I were having back and forth on Twitter ten years ago. Uh, mostly because we, we both shared this focus, if you will, on on who needs compute the most and was simultaneously had the least access to it and the shared interest and differences in approach. Earlier this year, it made me really want to take a deep dive into what Carlo and his crew had created with Nodeweaver. And man, my hair got blown back. Mike Matchett: Yeah, let's get into it. Okay. So what we're talking about with Nodeweaver, uh, you know, it's at the top level. There's a lot of people talking about VMware alternatives. It's not really about that. It's really it can be one. But we're talking about something where we're converging infrastructure, storage, networking, virtualization into a very lightweight system and really designing it for this edge use case where what you don't really have expertise, you don't have local management necessarily. So it's got to kind of stand on its own on there. Um, what would you, what would you sort of call that? If it's not a VMware alternative, where would you, how would you label that? Alan Conboy, Field CTO, Nodeweaver: Me personally. Go ahead. Carlo. Carlo Daffara, CEO & Co-Founder, Nodeweaver: Yeah. No, it's, uh, the, the, the key point is that the reality is that no one cares about the Azure infrastructure or the software itself. What they care about is running applications and probably running them in places that are far away or in many locations at once, Uh, with different requirements, uh, with, uh, difficult environments. And the reality is that what we do is infrastructure. You don't care about how your power comes through your socket. You know that there is a lot of it, but you actually don't care. What we are trying to do is to remove the, uh, pieces that you are required to know to use your application where the application needs to be executed, or where you have to collect data, or where you need to do your AI models or whatever. Mike Matchett: All right. It gets excited a little bit. Tell us what this looks like, what it's doing for us at this kind of location. What is if I if I deploy Nodeweaver to an edge, what do I get? Alan Conboy, Field CTO, Nodeweaver: What you basically get is a slick zero management edge platform. That's at the end of the day, it's hyper converged Infrastructure done the way it should have been from day one. Not only self heal, self load balances builds in the V switching, but has its own internal optimization adaptive engine that runs on not this specified little tightly controlled ACL or the ships as this kind of an appliance, but runs on whatever hardware either you have or you need. Period. Full stop and automatically over time optimizes itself for whatever gear happens to be underpinning that cluster at various points in time. In other words, edge infrastructure usable by anybody doesn't require the VCs and Cnis and Snia certified storage engineers, but rather. The. The. The guys on the ships can run. And they don't have to time delay themselves waiting for hardware to do it. Mike Matchett: All right. One of the one of the things that drew me to this, uh, is really, you know, something you've done, uh, and you mentioned here about being kind of autonomous, the thing runs itself and it configures itself to whatever hardware you already have at that location. Uh, so you don't have these big maybe setup costs. You don't have to deploy new sets of hardware that are cookie cutter. You don't have to go buy the latest and greatest, which, by the way, is hard to do today to actually go buy new hardware for various reasons of getting the GPUs and the memory and the servers that you want. It's getting hard to get the hardware. Um, so, uh, how does, how does, how does Nodeweaver do a better job of adapting to the infrastructure that's already there than, say, some competitors or things? Other things you looked at? Carlos. What, what I know you told me about P scores and stuff. Maybe you could just tell us a little bit about sort of this approach to, to, to using what's there to your best available, uh, opportunity. Carlo Daffara, CEO & Co-Founder, Nodeweaver: Yeah. The first thing is that, uh, any hardware provides you with an enormous amount of information about what it's doing and what is capable of, uh, things like, uh, the cache size, how many cores, what kind of cores? What is the ability of adapter to different temperature, whatever. There are hundreds of data points. Uh, the other thing is that instead of trying to optimize for every kind of thing, we basically run a sort of simulation. We tried to think, let's say that you want to add a new workload to your system. What is the impact of this on the rest of the of the infrastructure, on the other workloads that you have there? And we run this on every node that is part of your system, and we simply find the piece that matches Tetris, like perfectly with what is available instead of shuffling things around. And this takes into account lots of different attributes, not only the CPU, and it adapts on the fact that all hardware is different, even the same architecture. When you move to a new generation, you change a lot of things. So, uh, it's, uh, in a sense, it basically thinks, what happens if I do this now? And they collectively decide, okay, you are the best place to go do it. Mike Matchett: Yeah. So really kind of self-organizing Self-leveling if you want to think of it as self optimizing, uh, on its own, right? So yeah, so taking advantage of whatever infrastructure you have. And so that means in effect, you can build 2 or 3 node clusters out of completely different things and plug them together. And, and it will learn where each workload will run best or which will give you the best total experience out of that, uh, which is, uh, you know, I think if you compare that to say like, you know, this dynamic vmotion idea where something is going to be running until it sucks and then it'll move it somewhere else. And you basically just get this game of experimentation where you keep throwing things up in the air until maybe something works, uh, you know, and settles. But I, I'm never really sure. The system's never really sure what it's going to work. Right. So now you've got something that's, like you said, simulating and actually finding the best opportunities for a given workload and to have the minimal impact on the entire on the entire situation. Alan Conboy, Field CTO, Nodeweaver: You know, you said something there, Mike, that I think is under addressed right now. The ability to mix and match hardware in ways that nobody else really can do. That can't be overstated. It's amazing to me that I can take that quad socket Opteron, mix it with those, uh, 2015 circa Silver Prox, mix it with the latest, greatest ryzens and epics up at the top of my rack and all have a single cluster. That's awesome. But you said something just a moment ago to note. Everybody else out there that tries to do two node. Because let's face it, hardware ain't cheap. Especially today when six months ago you tried to order a couple of one new rack mount servers, and you're still, after the fourth delay, waiting for a delivery date two months in the future. Two node Carlo and team solved without needing a witness, without needing quorum. And they did it years ago. And one of the most unique ways I've ever seen. Using what I would personally describe as a combination of DFS file and object with full transactional tracking and a canary in the coal mine. Now, nobody else does this because nobody else figured out how Carlo and his team not only figured out how to do it years ago, but have had it implemented in the field for half a decade. Yeah, we talked about Carnival a little while ago, Carnival Corporation and their Carnival Cruise lines and all the other brands that they own. Every single one of those ships has two, two node Nodeweaver clusters on them, built from mixed hardware, right, handling 15,000 IP endpoints per ship every day in some of the worst conditions imaginable. An absolute recipe for split brain with anybody else. They've been handling it without a hiccup, good lord, since before Biden ever got into office. That should give you some context here. And and, and something that is huge to my way of thinking from both a business and technical perspective. Mike Matchett: Yeah. So you mentioned you mentioned a cruise lines. Uh, so shipboard and ships earlier shipboard situations would be one of those kind of remote situations that would really test anybody's, uh, stack. Right. If you put it out there and, and, and the sort of legacy answer or legacy approach, if that's what we can call it, legacy already, because you guys are the new, the new emerging tech, uh, would be to really create a standardized node and try to make every node identical and put 3 or 4 of them on there. Uh, and, and really just duplicate that, that footprint. Whereas you come along with this, this design approach. Carlo, and you say, look, you've got infrastructure, you've got at least two nodes. Put Nodeweaver out there and it will start to self, self optimize, self correlate, uh, and build that. I think, I think there's a lot of people watching this. We used to cover a lot of storage. Uh. Emerging storage technologies in the past, maybe you could just tell us a little bit about your approach to approach to data protection and storage, because I think it's pretty unique. Carlo Daffara, CEO & Co-Founder, Nodeweaver: Well, the basic idea is that, uh, the every storage unit is, in a sense, connected to the other. It becomes a single entity, what we call co storage. It becomes a single place and everything that you have to write in the system, it becomes a divided into small data objects called chunks, which they bring a lot of information along with them. They have their own transaction number, so they may be connected to other chunks. Uh, they have CRC because, uh, in many places it's essential to be able to identify data corruption, especially if you have things like vibration or complex environments. And you have an attribute called a goal, which tells you how many copies you would want to have of this information. This is a important I want two copies. This is essential life critical. I want four copies. And basically, every time you write something, the system recognizes that it has been written only once and autonomously say, oh, we need to make a copy of it right now. And if something happens, for example, you lose a storage unit. The system intercepts the fact that the storage is not available anymore and routes it to another copy of it transparently, so you never even lose a second of of transactions. And as soon as you add more storage, it will start replicating again and becoming again, reaching the goal that you want. And Alan did a bit of juggling with this a few days ago and it was surprised that it was you were able to to remove one, two, three desks and insert the new ones while the system was running its VMs. And basically it made no difference because that was designed to do that. Mike Matchett: Yeah. So you've got, you've got each node really mastering its own replication, uh, environment, looking out and saying, hey, who else can, who else can host some of my replicates or my N plus one for me? Uh, and if, uh, it becomes, it becomes self-organizing that way. I just love this, this a cellular architecture in a way, or a cellular automation. Uh, and I mean, you must think that's a lot of fun. Alan, put your machines through hell. Alan Conboy, Field CTO, Nodeweaver: You know, Carlo actually understated that on a two node cluster, just like we've been talking about. Okay. I literally removed seven of eight drives at various points. Shuffling, I replaced one of them entirely, rebooted a node, went into raid bios, and wiped the Raid config off of it. Turned it into four discrete drives and then let it boot back up. Cluster was fine. Workloads were fine. No corruption anywhere. Everything still ran it. That's just stunning to me. That's the kind of stuff I meant by blew my hair back. I know other platforms out there. You lose a drive in this node and a drive in that node and you're just done. Even though there are three plus nodes in a given cluster, it's a nonevent for Nodeweaver. And it's because of that autonomous cellular architecture you were just talking about. It's the ability for the nodes themselves to know exactly who's doing what to who for how many cookies at any point along the chain. Mike Matchett: Yeah. Yeah. Alan Conboy, Field CTO, Nodeweaver: The fact that I could do it with just a two node cluster and not even worry about split brain. Mike Matchett: That's massive and really suitable for the edge. I've got one more question for Carlo that goes up a level a little bit. So just philosophically, Carlos, you you this is clearly something that's going to be enormously important to those remote cases that you designed for the remote village, the, the cruise ship, the, you know, even, you know, if I've got 1500 retail point of sale stores in the, you know, across the continent, uh, it's going to be for that. But this isn't something that doesn't scale up. What's sort of the scale up limitations that you know of with Nodeweaver? How big of a remote data center, if you will, or cloud remote cloud situation can I build with with Nodeweaver. Carlo Daffara, CEO & Co-Founder, Nodeweaver: Well, for most of our customers, the majority of deployments tend to be small because we are an edge company, so the edges tend to be small. However, the nice thing about self-organizing systems is that they can scale up to with nearly no limits. We had one, uh, work that we did with the European Supercomputing Center, where they wanted to test how big we can make this thing, and we reached two petabytes and several hundreds of gigabytes across thousands of cores. And actually, the system was behaving exactly the same. We only had to adjust the the user interface because we never thought that we would reach a petabyte. So we had to make space for it. But that's basically the extent of what we had to change. Mike Matchett: A little bit of a Y2K issue there. We didn't make the field big enough for the amount of storage to storage. So we got we got going on here. We didn't think that big. Uh, I mean, it's fascinating that you have a, you have a design that doesn't really have a limitation to it, even in its, in its, I want to say ruggedized, uh, for, for those environments, which is, which is great. Um, so we don't have a lot more time here. I, you know, I know there's a very simple UI. Maybe we could demo that someday, uh, and take a look at how you manage this. Um, and that you've got some, uh, security built in. You've got some, some, some really, uh, I say, how would you describe security? This I want to say, I want to say blockchain ish, but, uh. Carlo Daffara, CEO & Co-Founder, Nodeweaver: Yeah, it's, uh, somehow similar. It's not based on blockchain for mostly for performance reasons and scale reasons, but in general, we designed it with the idea that security is a should be a constant property. You need to maintain security at all times. You cannot have a second where it's not secure. So we have a model that is based on rolling releases. We have very small updates roughly every two weeks and we send micro updates. That changes some some aspects of it. And it basically build security as a sort of an additional layer upon layer, which makes it also fairly easy for us to replace one layer with a new one if a new technology comes along. Mike Matchett: Awesome. Awesome. Uh, so I know bottom line here, uh, if you're looking at edge computing, uh, this is definitely something that should be high on your evaluation list. Uh, simply because it's doing so much for you, uh, and seems to be able to deploy anywhere with minimal headaches here. Alan. Um, if, or even if someone wants to get a little bit more information about Nodeweaver, take the next steps, dive in a little bit more, uh, pointing at your website, but is there anything specific that you'd like the people to think about? Alan Conboy, Field CTO, Nodeweaver: So you'll like this. Okay. Yes, of course, on our website, you can gather more information. That's what websites are for, right? But we make Nodeweaver and zero touch deployment just available from our website. Put that in context for you. My first ever let's stand up. A Nodeweaver cluster was literally no more complex than grabbing a USB key, grabbing the image download straight from the website, stabbing the USB key into a server that I wanted to be a Nodeweaver node hitting F11 during boot, say not boot from that. It literally built the entire cluster itself. Rebooted itself three times, deployed the virtual switching, grabbed the workload. Sitting out on an S3 bucket. Brought them down, stood them up, created them as templates, then formed the templates into virtual machines and started them. And after the second reboot, I just pulled the USB key and stabbed it into the next node. That's it. And that's kind of the hard way. Could have done the whole damn thing with, uh, DNS ops. Had some, uh, text records to my DNS server, then I wouldn't have had to even mess with the USB key or the F11. That's just the kind of inherent approach that we take here at Nodeweaver. Make it simple. Make it manageable from anywhere. The whole shooting match. And it's all right there on our website. Follow us on LinkedIn. You'll see some of this stuff firsthand. Mike Matchett: All right. I feel like we got to give Carlo sort of his last words here. Carlo, if you had to give someone a recommendation, uh, as to what next step they should take, and they've got edge deployments, what would you have them do? Carlo Daffara, CEO & Co-Founder, Nodeweaver: Well, consider that especially in this moment in history, being able to deploy an edge platform is something that is, uh, uh, important that can be done on whatever hardware you have available. You don't have to buy new hardware, especially given the price that it does have if, if you manage to find it. And the other thing is that an edge platform should be easy because it should be an infrastructure for your company. It should be something that gives you a capability to run the application that you need wherever you need them. And that's exactly what we do. We don't do anything else. We only do one thing and we try to do it as best as possible. Mike Matchett: All right. Which is which is a great thing. Um, I think we got to wrap it up here. Thank you Carlo. Thank you, Alan, for being here today. Uh, I will just say one more thing on your behalf. The licensing for Nodeweaver is this amazing per node plan. It's not based on the size or type of our infrastructure deploying. You just pay per a single flat, low price per node, which is also amazing. Uh, hope they can keep that price going, that price model going into the future for, for all the folks out there. Uh, with that, I will say thank you guys one last time for being here today. Check out Nodeweaver. Uh, it's got a lot of interesting things going on. You're going to be happy that you did, especially if you have some edge locations that have been giving you trouble. Take care.
