Hashing It Out
Hashing It Out

Episode 80 · 2 years ago

Hashing It Out #80-Golem-CEO Piotr Janiuk

ABOUT THIS EPISODE

Piotr Janiuk is CEO of Golem a worldwide, decentralized supercomputer that combines the computing power of every machine on its network. In the Golem ecosystem, you can loan out your computer’s spare resources to others who need the additional power to perform complex computations and tasks.

Links:

Sponsor

AD Music:Nihilore-Liberosis

Donate to Hashing It Out

Now into its work. Welcome to hashing it out, a podcast where we talked to the tech innovators behind blocked in infrastructure and decentralized networks. We dive into the weeds to get at why and how people build this technology the problems they face along the way. Come listen and learn from the best in the business so you can join their ranks. Welcome back, everybody, to hashing it out. As always, I'm your trust to your host, Dr Corey Petty. Colin is not able to make it today due to some previous engagements you couldn't get away from, but today we have an interview I'm excited about. I've been following this project for a long time. Think it's a it's a interesting and useful use case for blockchain technology and as a goal. Today I have a poter, CEO of goalm to kind of discuss where it's been, where it is where it's going. Welcome to the show. Why don't you start us off by introducing yourself and talking a little bit about how you got into space and what goal is? Hello, everybody, might let me see your goal them. So I've been in the product since the very beginning, since the inception of the idea in fact, and well back then it was mostly about it a room interim being the big change in the way one can utilize computing power. In a sense that makes smart contracts executable, meaning that you can buype us some lawyers and stuff that you usually need an intermediary for. And it was very interesting for us because we wanted to obviously, but then do some rendering without any intermediers. For example, if you have a seem to render, you would like to under this scene on a render farm, you need someone to collect your money and transfer it further. But back then it was the way we look at it. So it's even gave us the opportunity to in fact make the settlement without the intermediary, and it should be believe that's the way easier to do it, with its room and about telling intermediaries and without using Fiat money, in fact. So from the works perspective, we expected it to be fairly straight forward and we decided that perhaps it's good to share some computational power along and let people simply settle between each other so when appear to peer, the centralized matter. So the idea was back then was exactly like this. We believe that we can build a three side the prisoner market so that their requesters, providers and developers, and requesters want some job to be done. Providers provider resources, but then it was only about computational resources, and developers provide applications integrated with, gone and hosted by providers. So this is this was the starting point. It's emerged partially on top of idea of rutuamation of etherium. But whereas which amastion of wetroom is about keeping the consistent state and compute their voute. Well, from compute the state across multiple notes in the network and keep them in a constenus. Right here we wanted to utilize the power without having this largely constenus, just small scale, perhaps to notes consensus when collecting contributing results. So this would be the upper then, and we spent quite a while developing it. We learned quite a lot along the way, especially that we imposed our vision...

...of how people should compute, meaning that we assume this three sides to form a prisoner market. So this is in sense a constraint. Maybe it's it should be a prisonmer market, maybe not. So along the way we figure out that maybe it's better to users to choose what's the best for them. Is it a prisoner market or maybe some other sort of market? is a descent? Fully decentralized because it was imposed the network as well? But maybe some users want to use it partially centralized. We don't know. Maybe they want to have a service but would provide some sort of reputation source. Hard to say. But it should be user who is responsible for choosing what's the best for him her, and platform should allow users to achieve it easily. We shouldn't impose directly in the protocol any assumptions of how users may use the platform. It should be generic in this sense of well being able to choose what sort of resources they want to use, how they want to use and how they want to settle for them. And yes, this is how goal and evolved and the way we look at it right now is more or less along the lines I described. All right, so I at a high level. High Level, I'm a user. I would like to do some computation. I don't want to do it on my machine and I don't have access to compute clusters that would really do this. Instead, I could use the column networks a bit my job, pay for it, assumably with goal on or some R C twenty token or is or and then my my results were returned. Right is that? That's like kind of the high level, kind of walk through of what I'm doing as a user who'd like to have computation done. As someone WHO's providing computation, I have some compute resources sitting in my either like data center or home network, that I'd like to better utilize because they're sitting idle. I can then broadcasts, frees these copies compute resources onto the gold network and potentially make money by or make tokens of some form by taking those jobs from users, computing them, getting results in sending a back. That's that's the general idea here. So for the first incarnation of bill and this is true because, well, the mock, preduce, praywork is exactly what to describe in this computer distance realize money. But ultimately it shouldn't be only this. So we shouldn't limit ourselves only to the smart, produce pray work. In fact, people should be able to well. From providers perspective, perhaps it's true you have idle resources and would like just to make money or share them, maybe for free, because you will live in some idea. But See, I ushould be able to utilize your eye the resources. But from the requestor's perspective it's not assually extending in that task and getting book a result, because that's maybe a Serbus. Would like to post a web server, you know, says, of supersistent way, and you should be able to deploy it to the network and you should run over there. So yes, in essence, requestors should be able just to run any task, not necessarily only the task with one input and expected out as a result. It may be a stateful task and it should be able to run over a long period of time. Why? Why the change? I imagine process of building that first implementation, putting it out of production, having users, upgrading and learning, maintaining it, doing support tickets. You learned quite a few lessons, was it? Was it that process that made you decide...

...to revamp a lot of the stuff in your in your like feature improvements, or was it? Was it something else? It was mostly about lessons learned, because along the way we might some assumptions about how the computations should take place, about how they should be assured by the protocol where in fact they shouldn't be assured in any way, because we want to make sure that, for example, this integrity of the computation, which is very, very important. But how to assure the integrity? Well, we believe that the only way to do it is to verify the results. But we have quite a lot of different classes of tasks, but it's very hard to very fine. So if you had to the terministic task, then it's fairly straightforward. You can recompute the task on some other note and compare it bit by bit if it's different than someone's lying cheating. But if you have non deterministic task, then perhaps you can use some sort of metric to compare how to results are close together, which is the case with fendering. But ultimately there may be no way of comparing those tasks and your incapable of actually saying if what you get is the value result. So we try to solve this problem in general, whereas the better approach is to of all the two users who know with what kind of tasks they are dealing. So if software developer develops some sort of integration, let it be the rendering key, or she may decide to not necessarily utilize any sort of verification but simply resort back to some sort of reputation source and only calculate on the reputable notes in the network. This is one approach, but it should be up to the user to decide what to use, and it should be the users who provide the services to assist it, or algorithm, for example, for redundant reification as well. It shouldn't be built in the core protocol. Her Protocol should allow users to do it, but it shouldn't be important on them. Okay, so it. The main changes were removing the verification process from the core protocol and baking it into a layer above which gives the users more options on how they'd like to not only like not only what kind of jobs they can do, but also how they verify whether or not those jobs are done correctly by some third party. In fact, I would not call it even very far by the we go how they assure integrate with the controduction. Okay, and and then, like in your current vision, what are the options that are available to the to the user? What kind of reification as we have right now? We done the verification us we have or can have in this next month stone, reputable source of trust to centralized trust. This is well, this is an open problem. Could do sense well the regulation, but perhaps we can move step closer to Arus finding asolution or mix of the rope. Yeah, so how like our house, there are more potentially I'm kind of curious, like like goalm most recently came out with like the next version in production, which is a, in my opinion, a huge milestone and in a and the result of a tremendous amount of work and lessons learned, as you as you've kind of alluded to so far. You can you give it, give a brief snopster stuff, like what exactly that milestone was and what you're currently capable of doing with an ecosystem in production? Well, so, there wasn't my phone. It's clay and there is going to be the next big minstone, which hasn't been announced yet. Fully. Yeah, so if you refer to clay,...

...then I can resort clay. If you refer to this, neck my next mist on the different story. Let's step up. Starts claim what claim els and then we'll talk about kind of what's been deduch as the next past down. So clay is one of the inkern. Well, it's a version of gold as presented in our original white paper and it's enhanced in terms of what brass the previous version could provide. And the mine encouncement in the most important one is does API, so allowing users to integrate on their own and providing their own integrations in the map produced framework on top of Goling. This is the most important part because until clay there was blender integration on and a few other integrations, but provided by gold and factor, and it was very hard for developers to prepare their own integrations. One means of overcoming it was preparing wasn't integration, which would allow users to simply write programs in Web Assembly and just deploy to the network. Nevertheless, they were really needed to using wasn't to have this more generic approach. With clay and escape Pi, users are able to provide their own integrations on their own, withoute assistance of the gone factory. That's drastically the moment feature like what you could do at drastics better what you can do or what you can get done using the network. Well, in fact you can quite a lot using the network, provided that it's my produce. So if you can do something in my produced framework, your certainly well, not manserarily, but you should be able to do it using the clay. But it not maybe very straightforwork, because when you use your local machine it may be easier simply to speak your tasking to threads, or processes for that matter, and it's not that easy to split tasks and set prepare the set up, for example with prices and finally finmouts or using the other market where you pay for usage. In going, you have to be know a bit more about how the task her computer and that how long it may take for them to compute on different machines to prepare integration at the next milestone. This kind of not sure where you're calling it, but it's certainly a kind of revamp slash change in how things work, and you talk a little bit about that and what it is. Yeah, so, so this is this is the result of all the lessons learned along the whole way. So this is based on brand new architecture. So at some point more than a year ago, we decided to sit in a room and start to a quite board and discuss principles of all M and the architecturable and and the lessons learned need because quite a while, but we came up with quite solid understanding of what it should be, what the architecture should be and, yeah, how we want to build based on it. And the decision made right down was that gold shouldn't be a platform imposing any restrictional users. It should be a platform allowing users to easily locate resources, computational resources that they need. Not only competition of power. It should be displaced. It should be rum if they wanted, it should be CPU or Montples, abuse, which abuse whatever, and it should be easy for users to located. It should be easy to express what they're looking for. So not only I want to calculate it cutolic task with a timeout, for example, I may be willing to grant the VM with twelve. Course I want this Vm to be priced with the innire pricing model of whatever prising model. So there is a line...

...which to express the twelve description around language for resources and matching engine which allows offers and demage. Demands describe what the sources we're looking for and offers describe what resources providers provide to the engine matches those two and allows request present providers and demands and offers well much and find each other in the network. So this is it. It's from imposing some sort of market and market conditions. We stripped it to the protocol of finding resources and on top of it we added a few layers of obstruction that will be implemented by us, but ultimately they are based on architecture point principles, which means that they can be in fact implemented by anyone. We want to provide reference implementation. So, for example, in clay and in brass we had doctor containers, or was right. Now it can be anything that is ranging from bare metal to virtual machine is, through containers, even clusters. So we should be able to expose cluster of computers to the network as a powerful note, but it should be cluster that can be used as a cluster, not as a single computer. So this is it, and on top of it there is a payment API, payment obstruction layer that uses it. theium is GMT and its use for settlements and in different it's different from what we have in play in a way that it allows you to describe almost sributrary pricing model. So it will be pricing model for services. It may be linear, maybe any function in fact, and you can have time outs if you wish, but you can also resort back back, you can resort to some third party services which can help you either locate better resources or, as sure, give you some guarantees regarding the notes that you're going to find or, for example, they can attest some other features that those features attributes that those notes provide, for example, that they are genuinely processors with enough power that you should be able to run your computation. For example, someone can say that they have twelve, course, but in fact they can have, I don't know, hundred and ten whatever, or just utilize health half of them. And in fact this is not really looking for. But you should be able to easily specify exactly what you're looking for, and this protocol should allow you to much your demand with the offer, provided there is one on the market. So there are those slayers, there's muching of demands and offers, there's market which, in fact, after matching, distributes its between notes. There's obstruction, layer of execution but allows you to express in an OBSTAC way, many different resources and on top of there's a layer of payments and settlements, but everything is obstructed. It's reference, reference implementations will be provided by us about in fact, it's based on the full separation of those layers. I got a man so many questions from here. That's that's ambitious, as it seems to say. It's a it's a very large generalization of what you previously offered or sought to offer within goal. What's what's the projected timeline for the next milestone like for all...

...of this, for at least like you know, like DP's of all of this? That's a lot of really difficult things. Well, but we've been working only for quite well and also, I believe that is a couple of months, smoke, couple of months. What makes you think you can do it? Because you've been focused on it, like maybe somewhat more constrained previously, for the past, you know, five years or so, and you've been thinking about these problems this entire time. Is that why you think you can? You can do it and you have a solution for it that's apparently implementable within the next few months. Well, so, first of all, we don't want to don't try to solve problems which are too hard right now. So we solve problems that we know we can solve and we allowed other problems to users or two services. So this applies, for example, to the UBRICATION problem. And the second thing is that, yeah, internal, we already have a people so implemented, which means that it is pretty far down their own I mean the implementation. So it's not only the architecture, there's quite a lot of implementation already in place. So what do you see like kind of that that the main service offering of the network, with this vision in mind, as it just become kind of like centralized Amazon web services. I want this computer resource to do this particular type of job. Here's a market, here's a way for me to request that and find it at a given pricing model. I then initiate some engagement on top of that. Well, perhaps if I would say that it is the case, I would perhaps make the same mistake that we made previously by assuming that the market looks exactly as we envisit it. I think it's mostly up to users, and not only the requesters and providers, but the end users who can use those decentralized web tree like services, who can impose some sort of m tasks in the network that they may be willing to see something that they understand that we don't know about yet. So I believe it's up to them, not up to us. It's up to us to provide the platform and provide reference integrations and show them that external developers can easily integrate to the platform. But I cannot really tell if competing, well, it's calling it a distance whye a Muslin web service with something like this, is the right approach. I simply don't know. Yeah, I'm just trying to look at like maybe that long term, but like what, what's going to be useful in the short term, because, because we don't know organizations choosing to use sometimes this for persistent storage over long period of time seems like something they wouldn't do. And so you want to use the network, at least in the short term, while it's being vetted and and discovered, to use it for services that are relatively short termed or not necessarily mission critical. Like I don't want to host my infrastructure on the goal network. So what am I going to use it for? For sure, but HMM, I think that they are quite a lot of open source project in a sense, open projects that should be hosted on going. If you won't, for example, to have a back and or computation for your service, whatever it is. Then you may use other existing infrastructure, amazing whatever, or you can use goal and because it offers some features that you cannot get otherwise. So this may be a sense of censorship, resistance and based on decentralization or some sort of confidentiality guarantees, if and clakes are used, for example. But maybe someone implements a valid, fully homework encryption on top of Goling and this is something that you can use and...

...to locate easily. And Yeah, just as you can locate any resource provided in form of single note, you can have a group of notes, which is that even a MUZ on our other providers should be able to interflace with government provide your infrastructure through it. So I believe it's about the expressiveness of the platform. So if you there are certainly some niches where you have some needs that cannot be fulfilled by the infrastructure as it's right now, and perhaps you should be able to find something you look for using Golm. If you would like to, for example, just for test purposes, deploy small scale blockchain, you can rent a bunch of computers in a data center, but they are centralized. For sure. It's very hard to get it fully centralized and in Baling it would like to do such a test, you should be easily. We want click in fact be able to locate decentralized bunch of computers that can appuratetly scene we like environment, through bocting. Let's see, I could definitely. I mean not that. That alone gives me a lot of ideas on what what could be done at least with them, like devops and testing, and also think about like integrations with current infrastructure of providing services that do currently right now. Most of the etherium ecosystem, if it tries to provide a service, it usually offloads that service to subcentralized party that does a specific amount of education aggregation or provides indexes on top of rednose or things like that. And it's kind of the name of decentralization. But because the tech isn't there yet, we rely on centralized services to do things cheaply. Like can you can you imagine Golem starts to kind of take the place of those services or offerings of services? Either choose the centralize one or choose the golemn route. Definitely, definitely, and yes, I believe that more layer two solutions evolve or present the my actually need quite a little full odictional computer soon. Bob Where of Kent and having could go in with the bickon for such of the show. computations may may be kindy white. What do you see as as like your main road blocks, your barriers, like what's what's going to make what's going to be difficult? This is in like in the current landscape that you can foresee right now, because there's always going to be trips and trial cupulations that you don't foresee. But like what are you seeing right now is like the main hurdle you have to get over? There are a couple, I believe that, or two most important ones. The first one is peer to peer networking. In funding, there is absolute bullet. So yeah, we can handle it because we can resort to, for example, relays that would allow us to keep going peer to peer, but with some crontal points allowing for well routing traffic. But ultimately would like to get rid of them. And it's not really that but in terms of being centralized, because if you look at what I've described from the perspective of a service, not as relate itself. Then the relay may be hosted on goaling network as a service on goal network and being plaid by the users who go on network. So this, this in a sense makes it well valid argumental using those relays at some point. Well, and start with them. Ultimately, would like to have fully real life, free, fully working people, pre network...

...ork. Yeah, other's not with is. What is it currently? Is it felt on test PP or let P TP? Well, right now the going is built on our own solution. So, but we started with the B and it's so it's still maintains like a cauld delia like like networking discovery process. Yes, okay, okay. And the second hardle is the scalability of its room itself. So, yeah, we cannot do settlements to on the main blockchain because it's infeasible. Even in cleigm and will in brass. It was infeasible because we had to use budge transfers to make it cheaper. So we se infeasible in terms of gas cost, in terms of confirmation time. That's sometimes was way too long. So some sort of scullability would be required. We can build our own player two, especially just for the lamp and it would be fine. We can buys it, for example, online place. But ultimately would like to have working general solution that can be used with gone, without us having to provide something, especially for the first platform. But we know how to handle it, although it would be better if you wouldn't have to, yeah, and someone would just provide working solution. What are what have you looked? intives of our terms of kind of scalability solution specific towards Golum on a theorium made net right now, like there's there's so many options that have gone through introduction, high death in some cases, of like what we're going to do to scale layer two of the theory. I'm like, what have you tried? What if you what have you experienced, and kind of what are you thinking about, at least in terms of like analogous scenarios for calling? Well, if something like up to me to crow ups that scale would warm, then that would be just just fine. We were closely co operating. We'm using out some point and so, yeah, we wanted to use plasma, but we all did what we've found right now. Yeah, but the plasma was something that we're looking for, but in fact if you build a custom solution, for example, I use and place to close the get inside them and you just run gets the assigned again place, then perhaps it's wot it approach to create a side chain just for one project and just check point every block or so in the mine chain. I see, but can you, can you explain a little bit about that interface between is because, like the computation network, is definitely outside of the theorem network, but you need a link. That's usually through, what I from, if I'm not mistaken, payments and settlement of agreements between the different actors in the network. Is that basically what you're using it for? and kind of you describe maybe, like like the bandwidth requirements, like where you actually touching the chain and why? Well, so, we're we don't then products me. It's so we are not touching change. So back to play. Yes, we thought of the chain during the settlements, when the settlements happen in budges. So when you have enough payments accumulated, you send a batch of payments. It's cheaper than sending one payment after another. And this happens once once in a while. It will happen one in the one and that once a day pro you have, like you know. You know, you have your own infrastructure that kind of aggregates, it manages us as payments. Well, this happens on every note. So this is a fully socialized infrastructure. And then what you had, you foresee that changing as as things go, because like that they the more activity that you get, the larger your network grows, the more, probably an exponential case, transaction throughout your actually need. And whether or not and how you batch those things and aggreate them will to ask the determined, like whether or not your network and scale. Yeah, so right here we definitely need some lt solution if it's going...

...to scale exponentially. Done, any solution with fixed amount of payments that can be processed at time will make people an unusable at some point. So there have to be a solution that can take the load of transactions. And Yeah, it's cannot happen even on shrawdy to room with the exponential s calling soul. There has to be a very, very fast solution that can handle this amount of transactions because ultimately, if you look at this, yper someone when we not cycle. But in fact, if you have all in near payment model, you can play for every second of using the distributing payment so drastically increases the transaction required, transaction through. But yeah, exactly the so backing, I don't know, two thousand and sixteen, perhaps fifteen even with allly described first approach to payment channels exactly with this example. So, yeah, there has to be some solution external to the mine chain that can handle this volume of transactions. Do you see that as a tenement to you moving forward, like if that doesn't happen, you can't move forward, or do you just see, if that doesn't happen, to moving forward to the limited like feature set, limited options in terms of what you can pay, how you can pay? So so the best outcome from for us is someone preparing colayer solution and yeah, giving up to the world and well, I'm using it. If it doesn't happen soon enough, then we can come up with our own well solution, for example based on the place, but not necessarily. And if that, I don't see a scenario in which it cannot happen. But provided that we have some carvels, we still can wrong goal. I'm using the mind chain and watching or probabilistic payments. So it will work, but the better. Does call of the solution beas that the butter use user experience for the Blafferom, are you still married to a theium? Is that like say, for instance, a theium doesn't quite provide it, but someone else comes about and does. Is that something that you'd be willing to move? Is it? Are you attached to a theoreum or or what I mean? I it might at my point, like it behooves any platform to stay true to their initial at like I guess endeavors, but always be positioned to heads if they need to, because we're not guaranteed anything in this ecosystem. Yeah, for sure, but it seems to be the safe as but for us right now. And Yeah, I agree that. I certainly agree. But like we just don't know how things are going to go forward with kind of how F two played out or Fonet X. Yeah, yeah, if all those three approaches that they're described are not feasible, which is very, very highly unlikely in my opinion, then we might think about what to the next and talk to you a little bit more. Sponsor the show this week. Status and today I want to call out the many listeners who are building deps on a theoreum to tell you how to get your DAP in the hands of all the status APP users. Status APP itself is a mobile web three, Ms Your chat browsing transact there's a lot of cool things about the status APP. Right now let's talk about the DAP explore. Status AP USES DAP DOT PS. That's referred to as depts as an on ramp to use a Theorem daps on mobile. Maybe you've heard about defy. Want to check out Kayibra swap or defy zapp. Will get some ESSENC and F load it up in your status wallet and you just Daptot PS, DAP DOT PS to get defy on mobile. Take your decentralized, permission less finance with you. Already we're seeing tons of excitement around mobile depths and web three. If you've got a DAB, head the DAP DOT PS. Check it out, follow the instructions for staking, get your debt ranked and featured, or email steak at dap dot ps for more information. What's really neat about the status at DAB explore that it automatically creates a social channel for your DAB. You've got a...

...place for status APP users can find and use your debt, but also you've got the built in private, secure chat functionality to build a community. Do Qa Faq support or even mean building. What's that? You say you're not adapt developer. Why not? Status has a sweeter developer tools to get you started building, testing and deploying web three DAPS with embarked Io. You know, you see projects that raise a bunch of money in the night, seo in two thousand and seventeen, and then nothing. Some crappy wallet, maybe some Marketing Partnerships. But status is shipping consumer products, Dev tools and fixing a theorium and basic peer to peer networking and communication protocols. The teams legit. I'm on it to searchialize and open source. Check out everything they're up to at the status Networkcom or start with the status Apup at status. I am get that status dot. I am get back to the ship. So I have a server sitting behind me in the room behind me and it's basically idle. It's like a sixteen core server, thirty two kicks of Ram or so, and I want to use it. What what do I what do I do now to like get an online to do stuff? Or first, can I use it now is it a useful thing to even for me there to kind of like break it down, get it ready, start and submitted to the network? And what can it do? What can expect from it? Well, so yes, you can. You have to inst all the goal and client and we have roll through the configuration process, which is not very easy, but I bet you can do it. And once it's up, that that's it. It's runs and ways for tasks from the network. So from the provider perspective, it's installing the client and configuring it properly. Several mimerses it APP I need like a static Ip do I need? Is there any bandwidth requirements that I need or uptime requirements that I need in terms of like power goes out? Am I going to be? It's going to be an issue there. That depends on what you would like to do with the server. Because so, okay, start the gap. It's best if you have started. Gave you, but perhaps most of people do not. So, unfortunately, you need to follow the reports and configure your rooper accordingly so that, in the spirit of Pere not work on other notes can find you. So so this is the hard but once you're pasted well, it's really up to the profile of what would like to offer to the network. If you will know the what you can often turn off, switch off the computer, use it for other purposes than you don't accept tasks we've time out longer than some minutes or moll seconds, is to to the short but perhaps minutes. And this is it for now. But if you believe that this computer can be on for works of the day, then perhaps you can accept large talks, which are plenty. And the obvious thingment I think about, based on what I write do in my day job, is what guarantees do I have that the server that's running on the network in my house of other goal network? It's just doing arbitrary task that I said it to do, that allows to do. How do I know that the jobs that are being run on my computer won't infect that machine or the other devices on my network? Okay, how is it sandbox appropriately so that I have at the set a modicum of security guarantees? Okay, so right now you have a doctor container. We vote permissions to excels the Internet. So this is contend arized. And Yeah, this is the isolation layerror. And additionally, you accept task is the well, in fact, there are only a couple of different classes of tasks right now, which means that you won't be harmed by them. And you have the ones, which is a general purpose language, but we've conternalization and limitations in bosed to it's, I believe, very, very...

...hard to get out with. was from the doctor container. As far as I know, no one has done it yet, so there's no one hundred percent guarantee. But this is the security manners and as you expand, like I feel I'll sell your dress for expanding the capabilities of what you could do with a machine, which I feel consequently has consequences on what the machine is capable of doing to itself in the network. Is that's something that you're activity had or thought about? Yeah, you're right and yes, yes, well, of course this is the decision that it's up to us. So in the next monstone you should be able to for example, this virtual machines in the setup that you should understand or show someone else should attest that is secure for you as a user. So we've the set up with this virtual machine is secure and you trust that their parts were whoever who attests it, then you should be pretty safe from whoever is run inside the machine, inside the virtual machine. On the other hand, if you are experience user, a note you're doing, if you understand the tasks, you can even rent your burn metal and Exposit to the network. Oh Man, make sure you're the landswer set up cordingly. So I think I just I'm curious that it thinks I why. I find it fascinating on like how to do this stuff, but it feels as though anyone who can actively participate or these, especially in the next pasta, needs to be relatively savvy and tool and configuring, Lin existence their network, so on and so forth to get this thing up and running and participate the network appropriate. Is that your main audience and if you'd like to expand about that, like what educational on boarding procedures or you are you going through to try and get more and more people to participate without having to have, you know, deep technical knowledge? Well, so, yeah, initially I believe this is mostly for savy developers, not for regular yoursels, but only initially. People should get interest in the platform, should figure out how to co operate with different integrations, but they can put on the platform how to play with the platform in fact, and what they can take out by using it. But this is only an initial group of users. I believe that we're going to learn quite a lot by just getting feedback from them from later, further down the road road we should we should look on wh will reach out to other audiences as well, but starting with this specific adience in mind. So even for this audience it's important to have pacatons to get the generally be close to them, assist them and get feedbo from them. So we are going to organize meetings and, yeah, see what comes out and in the meantime I'm sure that quite a little those people's are going to come up with ideas of how to utilize going in ways that we haven't been an imagine and perhaps provide services or but perhaps services that can make the UX for regular resorts much better in terms of accessing the platform and making sure that what they run on their machine speces for them and in fact that what they what is run is, for example, running a computerial monitor. This is what the request or needs. Yeah, that's definitely like a y'all. Y are mainly based out of I know some of your based on a Perlant I've made major office in Berlin. Is that where the whole team is or how are y'all distributed it as a team? So the the major most of the team is located in Warsaw, Poland. Okay, yeah, and some of the people are in Berlin and across the world, but most of the depting is important in course. Okay, that's a good place to say. I have our our system advents and Paul and is. So I haven't.

I've kind of excided about this. Where people go? So I happens every time I talk to you guys, because I I've come from a high performance computing background. I don't I don't necessarily foresee high performance computing coming from a gold network, but there's so much general purpose computing that can be happened, that can happen that it seems like a reasonable thing to have such a network provide such services and people who have like me, you just have random computer resources lying around to participate in this network and passively our money. Like, do you see People's passive resources actually earning them more money than that amount of money it takes to run them computing? The I mean, like, is it you see them as it being profitable to run them a capacity, assuming the network is booming, instead of just kind of maybe breaking even or not being able to pay for that, like the Associate Electricity of running the jobs. This is very good and hard question. I don't sorry how if they work out with that, single notes and individual providers, because, yeah, right now, right this set up, you provide your note, which cannot be customly adjusted to the needs of the market under external world, and you need to trust that there is enough demand for the power of the type that you provide. Exactly. So this this is tricky. Perhaps yes, perhaps not, depending on the request ors and the yeah, they willingness to use such resources, because they may use resources that are cheap, but a children will look for something different. So this decentralization may be very important. Touch a good that request will be looking for, and it does scenario. Actually, individual notes, of yeah, single users should learn and make profit, but that depends on requesters on the other side. On the other side, they're providers with more resources, for example small clusters, and they can adjust the needs of the market and they actually, I believe, can make profit in an easier manner because they cannotjust what people require that's just a guess. But yeah, we've been thinking about it and it seems that for individual requests it's may be harder to easily get into do this market. But at the same time, yeah, it's just your ideal machine. So if it's idle all the time, you can still make some past passing profit. Well, income. It's not necessarily profit. They as next question there's like how to keep people to use it? Had you generate demand on this is good question. So, as I said, we're going to start with is this group of subvie developers and would like to reach out to other advances after getting a few iterations with them and seeing how they can interact with the platform one as the channel thing is, for example, have an integration with this platform that people right now understand. So there are quite a few projects, open source projects, which allow users to share their computing power to some greater dud, let's say, folding at homer right now, well, fighting covid so we have something like this, you can top of go them. Then it should be easier to uper users just to try it out. This is just an example. So, but those two are the examples of all we can reach out to auvious is some might more recognizable. That's definitely something. I feel as though. The more popular and effective those campaigns are, like folding at home, fighting covid things like that, the more you can leverage them. Right like that's folding at home has been around for a very long time. So it's like the Seti Project, right where you've used to use like the network of playstations idle time to search the stars for signals. I just think...

...that platforms like this can really leverage I have resources well, and everyone well, I don't know how computation, computational resources at home are going to trend over time as we move to a more like personal mobile development. That's me. People have desktops or servers, like I'm curious kind of like I just I'm form of the questions as I chat to myself, like have you thought about the personal trending of available computation for individuals at home, or are you expecting it to kind of move towards organizations, to small clusters who have extra resources, or individuals who have extra resources? So if I were to bet, I would, but on clusters, then in individual whosources, and the third stage is individual school just want to share and make some positive income. But that's guess, although I believe that's well, they are we can adjust to the market conducers, the more likely we're going to sustain. And, yeah, we go to this network that's going to have an effect on a broader ecosystem of quote unquote, decentralized tech. Right like there's there's kind of a few different paths that you can go down. One is everyone has a reasonably capable device always on and their home that manages a myriad of services for them, right like, I guess, what like reread x or good plus calls like the agent of their home and you can ostensibly see something like gold sitting on that thing and providing, providing resources for passive income. And or the other one is no one has devices at home and they relate and they delegate all computation to some third party, in which seems like a step towards set away from the ideology of etherium. I agree, and in the retrospected something like mind frames, but well, many years ago. Yeah. Well, I think that would diversified really, because there is point in hosting the machine in your own computer. We go to own place and and also they may be a point in simply a floading most of what you need daily do some remote machine and I think that I will don baton either being the winner, I rubber, but that those two are going to yeah, I guess they're all ill. I guess, like a that's a false dichotomy, if you will, because, like, it's not. It's not a dichotomy. It's really a kind of I want to call it linear, but it's a gradient of people's needs and demands and requirements of what they how they'd like to control things and how much trust they're willing to delegate to something else and the associated costs and available infrastructure for them wherever they are. So like a, because you're balancing all these different requirements and needs and availabilities, like you're going to end up with a myriad of kind of network topologies. Is that's like, and that's something that you're hoping this like your services are able, to a cost, back accommodate to like when guy request something, I can choose where it lives, what it's associated with, things like that. Like I'm maybe shifting a little bit towards that request ape that. Do you want to mention? That's a new thing. That's not something that I'm currently aware of in terms of, like, I need this, here's how I weary a network of things to get it. Yeah, so, yeah, so maybe...

...you are not fully capable of choosing something from the network because that's something might not be there, but you should be able to easily express what you're looking for and it gets there. Shouldn't be delivered to you. How far along is that? Is that that syntax? That like? That? That that language of for requesting things? Hm, we have, HMM, right now, I believe, two versions. The second one is the the full featured one, and there right now the p the verse is older, implemented and, yeah, ready to go. Even today we had the internal demo using it and see to work. Yeah, curious that up. I read up about that and maybe maybe bring you on again just to specifically talk about how that works in the INS and outs of kind of the difficulties there, because that's that's fascinating itself. Is like this queery API of a very generic set of resources that may be interesting them. All right, well, I guess we can wrap it up from here. Where do people go? Or first off, were there any questions that you would have liked me to ask, but I didn't. Well, in fact, I believe it was quite ter so not necessarily writing. That I wanted to mention in the meantime, when you touch the different topologies and different usages of the network, is that we are aware of at least two use cases where people have small plasters and those clusters are not necessarily clusters you got are designed to be exposed to the network as a full flashed Lata centers or sort of powerful notes, but they are there. For example, a company with an internal network, and those small companies may be, for example, in the Games and if you want to run a distributive compiler, on distributed resource processor, then if you are an indie studio, you may actually be limited and may need to grant some additional resources. So if you have such a small company, then from time to time you may need additional resources that are expensive, are and hard to get. But if there are a couple of indie companies that are connected with this network, provided that they really don't use this whole capacity all the time, they can actually interchange resources between them and use them to have their pipelines working smoothly all the time and the other one is simply scientific computing, where also you have a profile order of HBC or yeah, or you simply need a auster. Needed those from time to time and a police thing pull on the I believe that thing on other conferts maybe different. It's not usually easy to set up different clusters into connective each other and sense talked to each other. Now it's absolutely not. I could. I can tell you from a lot of experience that running a piece of software across a myriad of high performance scientific computing clusters is a nightmare because they're all configured differently and they all have different QU systems. Like it is the goal there to kind of marriage of the of the resources in I guess, like through a unified q because, like every age, every one of these HBC systems has their own que and kind of resource restrictions and allocations and how that q was managed in prioritize. Getting people to unify that type of thing would be very difficult but incredibly useful. But also like, because every single HBC cluster is very finely tuned in purposely built towards a specific type of hardware architecture, like a lot of at least...

...high performance computing a lot of those jobs. The software is also finally tuned for the architecture itself too, so like you can't just like submit a job in it, good in it and it uneasily across a myriad of different HBC clusters. So I like that. That would be hard to do, but if it could be done, I can imagine a a tremendous amount of theouse case. Maybe it's just like it's subsection to a certain type A certain type of job. Yeah, and I fully agree. And you can approach it to at least two ways, perhaps more, because you can either prefer a task as a request or of well, in fact the software developer, in a way that can be run efficiently on multiple configurations, which mean that this dosk inside would be comprised of a few different tasks that wouldn't be sent two different types of hardware on the other that site, or you can try to somehow unify the gateway between the network and the HBC to run the task. Yeah, but it's not up to us, it's up to users. If they believe that this is something that it's valid, use grace and I believe that someone would try to implemented problem with HBC. It's usually that getting people to understand and implement the assistem administrative starts like getting it getting a specific type of software optimized on a machine is hard to do, and so having them do this for very forward thinking type types of projects like column would be even harder to do. Business is, with my experience, maybe six, five or six years ago. But it's something that I think is well worth the effort. And there are are there are are subset of very popular scientific computing software that are amenable to such a use case. I look forward to seeing kind of that happened within the goal and ecosystem. Well, let's wait and see what's happens. Yeah, Whek and where can people go to learn more? Getting time contact with you? Start down on the client try it out for themselves. Well, you can go to a web page ontil. You will definitely find all the NASSARY information there. So go on. The network will route you also to our GITHUB, which is open source. Of course. There's quite a lot of source codes for anyone interested and quite a lot of documentation as well regarding configuring going, regarding how it works, rewarding. Well, not also not only going, but regarding additional technical information and additional information such as blow post, where you have described the different heard less not obstacles, different issues that we have to tackle along the way, and perhaps this something interesting to you as well, or to a user or developer. Great thanks, camera on the show. Thank you very much.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (118)