#341 From Google to Building Scalable Data Storage Solutions for the AI and HPC Era with Björn Kolbeck

[00:00:00]

Mehmet: Hello and welcome back to a new episode of the CTO Show with Mehmet. Today, I'm very pleased joining me from the US, Björn Kolbeck. Björn, the way I love to do it is I keep it to my guests to introduce themselves. Tell us a bit about your background and what you are up to [00:01:00] today and tell us also about your company.

Bjorn: Um, thank you for having me on the show. Um, I'm Bjorn Kolbeck, co founder at Quobyte. I started my career as a coder and then moved more into, um, the other roles inside a company. So I think I spend a lot of time in sales and marketing today. But my background is actually, um, software engineering. So I studied computer science in Berlin, um, then moved on to do my PhD at one of the HPC centers in Berlin.

Bjorn: It's basically a supercomputing center, um, that also did research in distributed systems. And that's where I met my co founder Felix. And within our project, the idea for, I would say the idea for Corbett was born. Um, we worked on a Or created a project called XtremeFS, an open source distributed parallel file system for grid computing, as it was called back then.[00:02:00]

Bjorn: And that was, in the end, the foundation on which we built Quobyte later. Um, before, before we started Quobyte, both Felix and I worked for Google. Um, that was an inspiration, a sense of how To run infrastructure at a very large scale and how to do it in software only. Um, and with this background, we decided to start Quobyte to bring this kind of approach to the enterprise.

Bjorn: So on the one hand, um, the idea was that HPC scale out workloads will come into the enterprise, which happened if you think about, um, machine learning as the latest technology. scale out workload in the enterprise. And then the other aspect was that we wanted to give enterprises the ability to run the infrastructure like Google.

Bjorn: So. Software based, very scalable and small teams can run really hundreds of petabytes of, of data infrastructure.

Mehmet: Great.

Bjorn: Little better in a nutshell.

Mehmet: Great. Great. I'll, I'll go deep dive also a little bit more, but just in a question [00:03:00] that you spark in my head. 'cause I, I, as I was telling you before we start even recording this episode today, I work in infrastructure for like, I would say like more than 10 years.

Mehmet: Um, and you know, whenever I used to see something cool coming out, especially in that domain. People like, it looks like Google is a, is a school where, you know, a lot of people, they graduated from there with these, uh, super cool ideas of how they can modernize, um, you know, the, the infrastructure, especially when it comes to the storage and, you know, the systems in general.

Mehmet: So you must have seen. Some I would say challenges that usually the enterprise face in the storage industry that you said, okay, like maybe we can, whatever we have learned, you know, from, from that school and also like the other available resources that were available to you at your co founder Felix. So what were like the main reasons that drove you to choose specifically to [00:04:00] solve, you know, the challenges?

Bjorn: Um, so my background is in, in supercomputing, high performance computing, and while let's call it an industry, um, is used to software storage. So, um, if you know the last profile system, it's all, it's all in software, it's scale out it. They always like the reliability part. So it was all about being fast, but not reliable.

Bjorn: Um, and that. Basically kills you at scale because the bigger your storage gets, um, the more failures you will have of hardware. So they always struggled with, you know, keeping the systems running, usually solving the problem with an army of PhD students because they're cheap labor. And then on the other hand, you have the enterprise storage.

Bjorn: Um, back when we started. Um, ExtremeFS that was in 2006 when we were PhD students. Enterprise IT was still, you know, VMware, uh, SAN storage, um, [00:05:00] capacities measured in gigabytes. Um, there is no shared storage. It's all geared towards transactional databases. Um, interestingly, I think not a lot has changed in enterprise I.

Bjorn: T. It's still very much focused on this kind of use case. And then you have the scale out use cases. First, it was, um, things like Hadoop. And now it's machine learning that don't fit the typical enterprise I. T. Model. And you just have this clash of cultures. So for us, it was clear that if you want to get those scale out workloads into the enterprise.

Bjorn: You'd need storage infrastructure. That's fast, like high performance computing, but reliable, like, you know, Google has it. Um, in the end, this mismatch is still here today. That's, that's the interesting part. People still think they can solve. Um, scale out workloads with a traditional storage. So, you [00:06:00] know, we started 10 years ago as a company, but people still hold on to those old ideas.

Bjorn: They're, they're slow to die, I would say.

Mehmet: Why do you think that, Bern? Like, I used to face this, I faced the same when I was sitting on that, sorry, on that side of the table, and then I became vendor also as well. So, why do you think Enterprise moves slow on the storage adoption, and then I think the innovation that comes there, although to your point, now you mentioned like machine learning and this new world closed, there is actually something reliable to your point.

Mehmet: But, you know, why is that? And how do you think we can change this culture maybe?

Bjorn: It's a very good question. So why is it, I think it's because, you know, when the car came, people said, why do we need cars? We have horses, um, you know, never, never touch a running system. There are so many, um, aphorisms out that it described the situation.

Bjorn: I think data is particularly. difficult. I don't know [00:07:00] who it was, but someone said, um, data or storage is the only thing that needs to work when, when you switch off the lights or the power in the data center can rip up networking. You can easily replace compute with new servers. But when your storage fails, as in losing data, you have a very big problem.

Bjorn: So people tend to be conservative when it comes to storage. Um, On the other hand, you know, you have been in infrastructure, keeping infrastructure running is often, um, you spend a lot of time firefighting and keeping the systems running. So the idea of wanting to change doesn't come from the necessarily the people that run the storage.

Bjorn: Um, but then you have the users. that are demanding new storage or they need it to run their workloads. If you want to really feed your GPUs, keep them busy for your machine learning, for deep learning, you need storage that's able to scale out and be very fast. So there's, there's a mismatch. And what I've noticed is that, uh, you could say the forward [00:08:00] looking IT folks, infrastructure folks, they definitely look into solutions to provide better, um, scale out to the users, but often it's, it's a bit of a fight.

Bjorn: Um, and that, that is, yeah, as a vendor, you're kind of in the middle. Um, you have to work with both. Um, but yeah, change is coming whether they want it or not.

Mehmet: Cool. Yeah, of course, like, uh, at some, you know, at some stage, people cannot deny that you need to do the change. Like, you can, you can resist it for a, While, but after that you need to do the change.

Mehmet: Now, coming back to the concept of software defined storage, you know, out there, there are a lot of maybe, and I'm sure they have faced this and it's normal to have, I would say kind of already existing market when you bring something new to the table, but. For you, what, what is like, for example, the, if we can call it the competitive [00:09:00] edge or where you saw, like what you offer would be beneficial.

Mehmet: And was that like specific to a certain industry or like, you know, we can generalize it to, to, to anything that can run these workloads.

Bjorn: Yeah. Let's start with, with the industries. I think, um, storage is everywhere and we cater to a broad range of, of industries. Um, our customers are in life sciences, media, entertainment, finance, um, traditional research.

Bjorn: So it really depends more on the use cases. So wherever you have the intersection of a lot of data plus performance requirements, um, we fit in very well and the drivers are different. So in life sciences, you have, for example, um, newer microscopes, they are automated, they have high resolution cameras, so they generate.

Bjorn: Tremendous amounts of data every day. Um, in media and entertainment, it's the high resolution, 4k, 8k video, where you suddenly need, um, HPC [00:10:00] infrastructure to do video editing. Um, in finance, it, it's a lot of machine learning, uh, autonomous driving. All machine learning. So it's, it's different drivers, but in the end, the underlying problem was always the same.

Bjorn: And then the software approach, this is where, um, we have to take one step back and ask about, you know, what a software storage, because software defined storage was so overused. So I heard one vendor say we are software storage because our engineers develop software that then runs on our appliances. So it's.

Bjorn: Um, that's, that was a very weird way to, to put it. And I think that's why we shifted towards saying a hundred percent software storage. Because for me, the beauty of software is, you know, you have a generic server. You download MySQL, it turns into a database server. I don't know, you, uh, run a UI and a web interface on it and it is a [00:11:00] workstation.

Bjorn: So the software defines what the server does. Um, an appliance is the antithesis of that. So software storage should be something that you can download and install and then turn the server into a storage machine. Um, that's what we do, but I don't think others are doing it. They have very, even if they claim their software, they've run on appliances.

Bjorn: Um, they tell you, yes, we're software, but you can only install us on this one server and this one appliance that you need. So, um, you lose the ability of, of software. That's so great to download and turn any machine into software, into storage.

Mehmet: So for you, I can understand you're like, you don't, I mean, of course there must be some specification, but you don't limit the model or right.

Mehmet: This is, this is the case now, I think, I think the. The issue here is the world storage by itself became so in the mind of people's kind of a [00:12:00] commodity. So people, they think to your point, like about it as, as it's just, they're buying, you know, a chassis and, you know, and it's just a commodity. And when I, you know, how they look on, on your website.

Mehmet: So you emphasize a lot about like, you are providing software only, which is, which is great. And I can see also like you can. Run on. I mean, anything at the same time, you can provide any protocol. Can you like, for example, like, you know, tell me a little bit more about maybe it's a little bit deep technical.

Mehmet: I know, but on a high level, you know, different use cases. So you mentioned we mentioned, of course, like high performance computing and you mentioned, uh, you know, machine learning, but If I want to put it in an enterprise grade solution, so can I cover all my workloads? If you can, like, walk me through, uh, the use cases from that perspective also.

Bjorn: Yeah. So, Quobite [00:13:00] is a parallel POSIX file system. So, it just looks like, you know, NFS or your local file system, which means you can basically take any application and run it on Quobite. Um, and then in terms of workloads, we built the system in a way that we can support the different workloads that you have, whether it's, you know, a lot of small files, which is a lot of AI workloads, uh, you need the latency for that, uh, random IOPS.

Bjorn: So basically databases or VMs that modify tiny blocks and overwrite them. And then also the throughput workloads like Hadoop or traditional HPC or video streaming. Um, And then we support flash and hard drives. Um, we decided that both storage media is important. Um, of course, flash is great for performance.

Bjorn: It's the modern thing. You want a lot of flash, but hard drives allow you to do things cost effectively. Um, there's still, you know, after 10 years, there's still five times cheaper. So this [00:14:00] combination allows you to then really run a broad range of use cases of Quobyte. And the idea is, Um, software or storage infrastructure so that you can basically from your core byte cluster as an admin, you can provide your users with different kinds of storage, depending on the needs.

Bjorn: So you can do the high performance scale of storage that they need for their workloads. At the same time, you can provide cheap and deep archiving, um, cold storage on the hard drive tier. And when you can. Deliver different types of storage from the same storage cluster you get great efficiency because you basically have to manage one system only You can share resources We both provide with things like multi tenancy over subscription and so on.

Bjorn: So in the end you can provide Storage as a service, like on the cow, very fast, different use cases, very efficiently. I think that's, it's, it's two or three steps away from where people are [00:15:00] today with appliances that are very focused on different use cases. But in the end, this kind of, Storage infrastructure is what you want as an admin because it makes your life much easier.

Mehmet: Absolutely. Now you mentioned something which I was trying to ask you about. So you mentioned the cloud. So someone might ask, you know, okay, but doesn't the cloud, you know, hyperscaler, they provide me the storage. So why still I need to use Uh, your solution, for example, and of course I know the answer, but I'm asking this for the audience.

Bjorn: Yeah. Um, it's a very good question. Um, the first one is that if you want to run scale up workloads on the cloud that use a file system, most cloud providers don't have a good answer. So we have customers that use us on the public clouds, um, and they love the performance and scalability there. And the other aspect is that the cloud quickly gets very expensive, especially for data.

Bjorn: So if you have, let's say, I think the tipping point is somewhere around half a [00:16:00] petabyte, the cloud suddenly becomes very expensive and then running on prem with the right infrastructure is actually, I would say it's the better choice from a cost perspective, but also from a. Perspective of keeping control of your data because that's the other thing.

Bjorn: Migrating petabytes of data is difficult, and it's a huge lock in for cloud providers. So keeping your data on prem under your control is a huge business advantage. Absolutely.

Mehmet: And I can imagine like also migrating. You know, the workloads would become much easier, right? Am I right?

Bjorn: Yeah, I mean if you uh, so one thing you can do when you have your data on prem is move parts of your data onto the cloud for temporary Um, clusters, it's basically, you know, ingress is free, copying out of the cloud is expensive.

Bjorn: So a pattern that we see is like you start temporary, or you spin up a cluster on the cloud, for example, because they have to key [00:17:00] resources available. You create a temporary callback cluster hydrated with your data from on prem, run your, um, deep learning, and then copy the model back and just tear the cluster down.

Bjorn: But monoliths. fairly small. So the egress fees are low. That that's a pattern that we have seen.

Mehmet: So also to the point of scalability and, um, you know, performance per meg, I believe this becomes crucial, especially in just like the use cases, as you just mentioned. So you talked about like life sciences and so on, and now everyone talks about large language models and they talk about AI.

Mehmet: So. If we want to put that into kind of some examples and or like some, maybe metrics, um, for the people who would manage probably, of course, the, even the AI and team, like the data scientists, and these folks, they should also give their feedback. But if I want to put that into context, so why here scalability [00:18:00] and Performance became key, and you need to have this approach that you are applying today to have the best results.

Bjorn: Yeah, it depends a bit on the models that you train, but in the end, um, you know, it's the old story of scale up versus scale out. If you, if you have models that you can only, train on a single machine, then it's a question of how many GPUs can you squeeze into a machine and then you quickly end up with, you know, there is an upper limit and that's it.

Bjorn: So what people do is they scale out. So they train smaller models or they split the models up and train across a large number of machines. And now The challenge is that let's say you train your model across a hundred machines. The data from your storage needs to get into those hundred machines very fast.

Bjorn: Um, because the GPUs need to keep on loading the data. Um, so with a hundred machines, you suddenly have. You're going broad instead of one machine trying to do it very fast. You have a hundred [00:19:00] machines that need to get it in. So your storage needs to be able to scale out and serve those hundred machines how so one thing that seems to be a challenge for storage admins is that you're not asking for the lowest latency.

Bjorn: It's not about, you know, delivering. A billion IOPS to single machine, you know, the old transactional database approach instead, it's about delivering consistent high performance to a large number of machines. That's the beauty of scale out. And then your storage system also needs to scale out and basically, um, be able to deliver that.

Mehmet: Now, to this point, and you know, maybe I'm going back to the concept of eliminating the hardware from the formula here. Um, do you provide also, for example, some best practices? Like, do you help the customers? Of course you don't sell hardware. You are a software company, but of course you touch [00:20:00] the heart of the customer in a way.

Mehmet: So do you provide also like best practices for them? Like, for example, based on the applications that they're going to use on the storage. So you prefer, for example, to have these models, do you do these kinds of things per there?

Bjorn: Yeah, we have what we call recommended hardware, um, because we don't really prescribe which exact models you need, but we have, of course, there's, you know, the high performance tier, all flash servers, um, then we have a, Mix of hard drive and flash or the all hard drive servers.

Bjorn: And you can get them from any vendor. I think this is, this is our approach where we say, you know, you, you know, vendor X, you know how to manage those servers, then go buy those servers. Or if you want, you know, The best price. Then you let the hardware vendors compete and asked two of your favorite vendors for a price.

Bjorn: Um, in the end, the advantage is that, uh, the margins in the hardware business are not great. [00:21:00] And we let the customers take advantage of that because the street price for servers is so different from storage appliances. Um, that's true. That's why it's an advantage for us to stay on the sideline and say, this is the recommended hardware.

Bjorn: Um, the other aspect of course, is standardization. If you. If you've run your data center and you can standardize on less different server models, you make your life easier and you save cost again. Absolutely.

Mehmet: Now how this affects, um, you know, like if nowadays everyone talks about saving, so does this help the customer also save money when it comes to, you know, maybe support different hardware to your point?

Mehmet: And also like, How do you see yourself in? Because everyone also talks about now, uh, yes, you know, initiatives and sustainability and all this stuff. So do you see [00:22:00] yourself helping customers with their concerns about energy consumption and sustainability in data centers?

Bjorn: Um, so for for the serve savings?

Bjorn: Yes. I mean, there's nothing better than commodity hardware because as I said, it's a highly competitive market. And so the customers benefit from buying the hardware from that market. Um, the other aspect is what I mentioned with the standardization. Um, we see that a lot these days. If you can reduce, you know, your data center to three different server models, um, your management becomes easier.

Bjorn: Which reduces your cost. And then also you buy a larger number of those servers, which gives you better pricing power. Um, with the, uh, green, um, or energy savings, I think this is where picking the right hardware can help. Um, again, um, you know, if you talk about AI, [00:23:00] uh, your, your GPUs will dwarf any storage.

Bjorn: Um, so we can kind of say, you know, save on the other side, but then, um, well, the beauty about scale out is that instead of going with the fastest hardware in each server, you can use servers that are lower powered and then just scale out and have more of them. So if you want to be energy efficient, you can pick servers that have energy efficient CPUs.

Bjorn: For example, you don't need to go with the fastest ones. Um, I think the challenge then is, is NVMe drives. You definitely want to use more hard drives because surprisingly those are more energy efficient than NVMe drives.

Mehmet: Yeah, true. Now we mentioned AI and of course we were discussing AI from, you know, AI utilizes the storage.

Mehmet: But have you seen any use cases where AI and all, you know, the buzz that's around now affecting, I mean, the main functionality of the storage itself? [00:24:00] I don't know, like, because, you know, back in the days we used to hear about, you know, the auto tiering, we used to hear a lot about, you know, uh, some kind of intelligence in the way how you do priorities for the applications.

Mehmet: So, are there, like, any advancements and are you planning or maybe in the phase of doing something that, of course, when I say AI, just for the audience. Don't think chat GPT. I'm not talking about this. I'm talking about AI and machine learning in general. So any, any specific use cases here?

Bjorn: Of course, we're looking into it, but I think in the end, um, when it comes to infrastructure, a lesson that I learned is that, um, you need predictable behavior from infrastructure.

Bjorn: Um, a big problem back in the days was, That's called distributed hash tables. Anything with hashing where an admin couldn't really predict how a system will behave. So I think that with storage, I don't know if people really want AI to change how it [00:25:00] behaves, making predictions, alerting those kinds of things.

Bjorn: Yes. But you want your behavior to be very predictable. Um, so item. I think, yeah, IOPS was a big topic. Of course, the usual suspects like Gartner's on trying to push that, make it a big thing, but I don't think that IOPS is going to be exciting. I haven't seen anything exciting in the past couple of years, to be honest.

Mehmet: Maybe another use case that, you know, just came to my mind, and this was also kind of a differentiator. Now, it's not anymore, where You know, the hardware can itself detect that, for example, if a disk will fail or, you know, I'm not sure like in this space. You know, how things can go into the future.

Bjorn: Yeah, I know this is, that was also my thought, you know, predicting failures early on being faster than smart and whatever with AI sounds good.

Bjorn: But then [00:26:00] we, we turned that around anyways with our architecture where we, you know, hardware fails, it's, it's just a fact, and you can try to make hardware more reliable by trying to build redundancy and whatever, but that was all expensive and doesn't work well in the end. So. We built Quobyte on the assumption that hardware will fail and failure is okay.

Bjorn: So we don't really need to predict failures. We, we can react to failures when they happen. That's yeah, makes sense. And then, you know, you don't need because smart for years was smart tells you the drive is going to fail. And then right before it fails, but this trying to predict phase early on, I know that I remember 10 years ago, I'm ready.

Bjorn: There, there were people trying to put microphones in the data center to identify failing drives based on the sounds and all of that.

Bjorn: Yeah, it's too late. The moment you know that something is going to fail, it's usually too late. So you have to build the [00:27:00] system on the assumption that everything fails all the time.

Mehmet: Yeah, so this is where you need to focus on how you have like the better, I would say, fault tolerance, the highest fault tolerance instead of focusing on these things.

Mehmet: So to your point now, you mentioned a lot burn like the what you call it the scale out architecture. You know, I walked, you know, in multiple times, both on both sides of the table and scale out architecture is really like, you know, whoever invented that is genius, right? So But, uh, people always used to question one thing.

Mehmet: We say it's a scale out architecture, but are there any really now, and you come from a computer science background and you have a PhD. So from scientific perspective, when we say we can scale unlimited, is this something real or are they like, maybe as we, we can say, like, it's like, it's physics, [00:28:00] like law of physics, like after some time we need to have like maybe another cluster or something like this.

Mehmet: I'm always curious to know and, you know, hear it from, from someone who's into the domain and, you know, who studied this deeply. So what you can tell us about that?

Bjorn: That's a very good question. Um, so There are, I think that in distributed systems, we showed researchers was always this, if it scales to 16 nodes, then it will also scale more.

Bjorn: Um, the question is really, where do you get the diminishing returns and what is keeping you from scaling more? Um, and then you have to ask, is it a, is it a fundamental limit of the architecture that keeps you from scaling? Or is it something in the implementation? So some, one thing we noticed is that when we have clusters that have, um, 250 nodes or more, um, we noticed that our monitoring became a bit of a bottleneck.

Bjorn: Okay, that's not fundamental. So we solved that problem. Um, it's a different story when you have [00:29:00] typical example is a, is a centralized lock service. Whenever you have resources that need to negotiate who's responsible for something, you typically use something like a lease or a lock. And with systems, um, that have like a lock service, you always have the problem that of course, everyone needs to talk to lock service.

Bjorn: So they try to make it hierarchical, uh, hierarchical lock service with similar problems. So the question is, do you have something. A component that inherently limits your scalability. Um, another example is my favorite NFS. The protocol is designed for clients talking to a single server. Um, when you have NFS in your architecture, you always have those bottlenecks of clients having to go through one or more NFS gateways.

Bjorn: So that's a fundamental problem. Um, and if I look at our architecture, um, we built it in a way that we avoid, for example, central lock service. Um, [00:30:00] that was part of my PhD thesis, how to do that and appear to be a fashion to make the system, um, truly scalable. Um, the other aspect was that we said, uh, no to NFS.

Bjorn: So the Quobyte clients where you consume the storage directly talk to the storage service. So there are no bottlenecks in between. Um, and then the third aspect is like how you deal with scalability, because, you know, there, there needs to be a starting point that everyone talks to, and then the system fans out.

Bjorn: And this is where we, I sometimes call it a federated approach, you know, like a, like a federal state where you have on the top, you have the federal government, and then you go down to the local level state, local. And this is how you make. something scale. Um, we have a similar approach where we have different levels and then we can scale really big on the lower levels.

Bjorn: Um, so far customers haven't broken that [00:31:00] in terms of scalability. It's still scalability.

Mehmet: I think what helps also, and you know, Just for, for people to understand. So, so the data can grow. So the size of the data will grow and you need like more, uh, more storage. And sometimes you need more, more compute also as well.

Mehmet: So the thing that also I think it's helping this. It's a sustainable architecture because computers always, computers always getting better. I mean, it's getting faster. And even the hardware itself from storage perspective, so you can get, you know, more dense from capacity perspective in smaller, uh, form factor, if I can call it this way.

Mehmet: So maybe that, I'm just. because you mentioned the appliances. So maybe if an appliance used to be like giving you a capacity 10 years back of 10 terabytes, now we can have the same size appliance with like maybe the same number of compute nodes, uh, 100 terabytes. I'm just giving, you know, making up numbers from my mind.

Mehmet: So I think this is also helping actually this architecture to sustain [00:32:00] because always You know, the compute and storage power is also growing at the same time. They are scaling by themselves. If I'm right, I hope I'm, I give a right example here, right?

Bjorn: Yeah, there is a degree that things are, um, getting faster, but there's a caveat.

Bjorn: So for example, hard drives aren't getting faster. They just get, you get more capacity on a hard drive, but it's still bright. 100 megabytes a second. Um, with NVMe drives. Yes, they get. More dense, but then, um, the power consumption is now a problem with the large NVB drive. So, um, it's, it's a bit tricky. I think the power and cooling is still something to keep in mind, but the good thing is when you have standard servers, you can do things like.

Bjorn: Distribute your storage service across all your racks, which is also great for reliability. But then, you know, you have storage in every rack, you might have balance of compute. Um, you get the flexibility there, but yeah, density is [00:33:00] definitely, um, I mean, if I, if you look at, um, those, those servers where you have, uh, four servers into you, that's actually a highly reliable storage system with high performance into you that you can put everywhere.

Bjorn: Um, that that's pretty amazing if you think about it, edge use cases, you know, just put it in the data center or in your, in your, I don't know where, where you have the, the it rec in the, in the store and suddenly you have a highly reliable storage system where it can do machine learning in the store. Um, yeah, the, the, the software, the hardware, the hardware definitely helps us.

Bjorn: And also the decline of, of hardware storage media prices.

Mehmet: Yeah, so I got it, Bernadette. You just try to solve these, I would say, bottlenecks on the software level and then the harder become just really now the communities. It's not like the place you focus on. And you know, for me also [00:34:00] myself as a personal opinion, I'm not big fans of, you know, the gigahertz and IOPS and all these.

Mehmet: Because even when you think about it from benchmarking perspective, these numbers don't tell us much usually, like they are just numbers. And even when, when we see some comparisons, okay, so what, you know, like the question that the customer might ask, okay, so what, like how this amount of iOS would affect.

Mehmet: And, you know, to your point, like the important things are to focus on scalability and performance in the way that you discuss it is to minimize What do you call that? The bottlenecks, right? So I'm the percent on this.

Bjorn: I couldn't say it any better. Yeah, exactly. It's like, you know, people still think in those terms of, Oh, I need a billion IOPS.

Bjorn: Um, but that's not the point. The question is, can you, can you do it in small units? You know, if you do a billion IOPS across a hundred servers. You can use cost [00:35:00] effective hardware. And the question is, does your storage scale out to those dimensions? You know, we're in a scale out world where it's more to aggregate.

Bjorn: And you don't think in terms of those crazy benchmarks, we try to squeeze the maximum from one machine to have the best benchmark ever. Now it's really the question. Can you do it? Can you, can you have a storage cluster of a hundred, 200, 400 machines that delivers, you know, this, you know, when you go broad, it's like a river.

Bjorn: It's like a massive. Wide river that transports tons of water per second. Can you build something like that? Does your storage scale out to that? If not, then you're lost in the scale out world.

Mehmet: Absolutely, and just again, sorry, I'm giving examples because, uh, it's like, I was telling you, like, I work in this domain for a while.

Mehmet: So, so you have low physics, which are outside actually the storage. So to your point, you just mentioned the hard disk, uh, throughput. And also because we still use applications that Client server applications. And then you have the network part as well. So if you network [00:36:00] assets, it's 10 gigabit per second.

Mehmet: Like what? You know, like I have that bottleneck also as well. So 100 percent on this. Now I want to shift gear a little bit because you mentioned something that you shifted from being a pure technical person to going out and doing sales marketing. So I would like to know a little bit like how was the experience?

Mehmet: How was Did you face any challenges or because you were like so passionate about what you were doing, you find out yourself like very like, as you were experienced for real in doing this?

Bjorn: Oh, there were a lot of challenges. Uh, I mean, the transition is never smooth, but I did realize quickly that I love being in front of prospects and customers.

Bjorn: It's exciting. You know, I think founders should talk to their customers and to their prospects or you know, to the market. And again, I've always get a bit suspicious when founders hide and just do, you know, the technical part because in the end you have to listen, you have to understand the market, you have to [00:37:00] understand the needs, you have to understand how the customers work.

Bjorn: And then, um, so that took me some time. I had to learn, you know, I learned from from our salespeople and I learned a lot. Um, how You interact with customers. Um, sometimes you just say no, you know, it's totally fine. Or sometimes you tell them, Nope, this is not something we're good at. We've not a good fit, which hurts at a founder, because of course the software, the product is your baby and you want everyone to use it.

Bjorn: Um, but I think for me it was crucial to be in front of customers and I still love doing it, you know, and then once the customers in production and you actually see them use your product, something that you worked on for a long time, it is also very satisfying. Um, the other part is that I, when it comes to marketing, I think I had the typical techie arrogance.

Bjorn: I thought, you know, great product technology. Oh, it's easy. Um, and I had to learn the hard way that [00:38:00] marketing is, it's an art. Sometimes it looks like a black art, black magic. Um, but it is, it is not easy because it involves psychology and, um, it's a lot of trial and error. And so I had to learn a lot about my marketing myself to be able to.

Bjorn: You know, understand is my team doing a good job? Um, should we try something else? You know, who do we need to hire? Um, Marketing people are good at selling. So of course when you interview them, they know how to sell themselves. That's not, you know, criticism It's just um, how it works. So you have as a as a family, you have to understand marketing And more detailed to understand really do they know what they're talking about?

Bjorn: Do they know how b2b marketing works in the post covid world where people don't really? work from the office anymore where, you know, those kind of things. Um, so that [00:39:00] that was a real chance I had to get outside of my comfort zone, which was coding, um, a lot of times.

Mehmet: Yeah, absolutely. And, you know, I'm, I'm, you know, one of the people who believe Uh, in, uh, you know, going out of the building when founders come to me, I said, okay, you need to go out of the building because, of course, now, from some perspective, you know, people ask me, but how do you know this?

Mehmet: Because again, when you work as a technology consultant for. A vendor in the tech space. So you have the same assumption. Oh, I have the best solution. And, you know, the customer, I mean, I can't be wrong. The customer is wrong. And sometimes it's not the case. So, you know, uh, and, you know, uh, it's good. You mentioned also about, uh, You know, I loved your transparency because you mentioned about, uh, I would not call it arrogance.

Mehmet: I would call it like being a little bit stubborn in opinions to make it a little bit nicer. [00:40:00] And assuming like, okay, maybe these people are not getting, or they are not the best. But for you, you know, like you said, like you had also to, to, uh, change the product to that, uh, uh, to, to what the customers want or maybe what the market wants.

Mehmet: So this is really good advices. And if I want to ask you, like, if someone today is a pure technical background and, you know, he, he wants to go out or she wants to go out, like what other, like, I would say advice you would give to someone who want to become an entrepreneur and start their own companies in the tech space.

Mehmet: Of course,

Bjorn: I think from my personal perspective, it's immersing yourself in the different areas of the company. You know, it's Technology is very important, but it's only one part. As I said, sales, marketing, you need to get an understanding of it, and also an appreciation, and you need to find the right people to complement you there, because you will never be the expert.

Bjorn: You're the [00:41:00] tech expert. You need good people, um, to complement your team there. So I think that that's really what I would say, from my perspective. Should have maybe done it a bit earlier. Um, Immerse myself in marketing, for example. It's fascinating. You know, it's, it's Once you look at it, the problems of marketing are fascinating because it's all about people's psychology.

Bjorn: It's the opposite of tech. You're dealing with Irrational human beings and try to convince them that they're not Your, your technology is the best. Um, I think, you know, after leaving the comfort zone, I appreciate how, how much of a different kind of challenge it is. I think that's why I consider myself very lucky as a founder that I have the ability to immerse myself in different aspects of the business.

Mehmet: And I think you start to get it more like when you were on the other side. of the table and people used to pitch for you something. So I think you start [00:42:00] to understand more. Oh, now I get these people where they were trying to do right. So, uh, because I lived this feeling when I shifted from one side to other side of the table.

Mehmet: Um, as we're coming to an end, like any final thoughts you want to, to tell us anything, maybe I didn't ask you and you want to highlight about, uh, go by and where people can find more about you and the company.

Bjorn: Yeah, I think you covered, you know, my favorite topic and I can't believe that, you know, it's, it's 2024 and people still try to wrap their head around scale out.

Bjorn: Um, so thank you for covering that because I think it's, it's really important that people understand that scale out is a different world. And this is where it is heading. Um, in terms of finding out more about us. Um, yeah, we have a free edition that we put out there so people can actually see that, you know, we really mean that.

Bjorn: But. Software storage can be easy. So you can download that from our website. If you have servers on the cloud and use it, it's free, it's free for production use. So [00:43:00] go ahead, see that, you know, storage can be something that you can download.

Mehmet: Great. I will make sure that to put the website in the show notes and at the end, thank you very much for being with me here today.

Mehmet: I love to talk about deep tech, of course, like anything that has deep And B to B tech, of course, uh, that people they don't see and this for the audience who maybe say, Okay, what's this topic? So anything you run today is trying on a storage. And, you know, I'm passionate about any technology that can make us the end users life.

Mehmet: It's easy in a sense. It's scalable. I know that it's reliable. I know like I would not have any issues when I run my application. So really appreciate the time here today, Bernard. All the nice things you have done, you and the team. Thank you very much. And this is for the audience. So if you just discovered this podcast by luck, thank you for passing by.

Mehmet: I hope you enjoyed it. If you did so, please subscribe and give us a thumb up and share it with your friends and colleagues. And if you are one of the loyal. [00:44:00] Audiences all keep coming to me and send me their suggestions and feedback. Thank you very much for doing so please keep doing what you are doing.

Mehmet: Thank you very much for tuning in and we'll meet again very soon. You, bye bye.