DAPPCON 2019: 10 Questions for the Golem Project - Aleksandra Skrzypczak (Golem)

DappCon 2019 · August 2019 · 22:10

At DappCon 2019, Aleksandra Skrzypczak of Golem answers a rapid-fire set of questions about the Golem project, a peer-to-peer marketplace for renting computing power built on Ethereum. She covers how the network matches requestors with providers and where the project fits among Ethereum-based infrastructure for decentralized computation. The format gives attendees a condensed, direct look at Golem's progress and roadmap.

Speakers

Topics

From DappCon 2019, hosted by Dappcon Berlin.

Transcript

so hello I'm Alexandra I'm acting CEO of the golem Factory and I'm here to answer some question which we have been asked many times before so first look at know you how many of you have heard about the column headed before okay and how many of you have use it really no that's I really hope that I will be able to encourage you to try it on final so let's find the first basic question so what is gone.we one of the oldest team gaps actually we have the first presentation about the

idea at difference you know in 2015 and then we got about 100 in 2016 which allows us to scale up the team significantly to be able to be able to actually build on the idea that we had in our months so what's the idea the idea is to connect all the computers across the globe to create one global market completely power it's not a new idea but basically during the Victorian twice people are trying to do something like that and to add the market to the

tolerance but what was missing during that time was this distributed consensus later when Bitcoin was born the distributed consensus for payments was finally in our hands and tighten it even we got the consensus that was not founded on the payments they earned the only statement logically so having this one additional component we finally had tools that allow us to make this vision a reality so to sum up there are three types of users in the column network the first group are requesters requesters anyone it needs computing power it can be engineer it can be a silent

it can be a blockchain validator that requires a huge amount of additional power the next group are providers provider that anyone that have unused computing power so it can be an individual user that has a powerful PC or it can be a small data center the distilling for additional possibilities it can be unused may include for example and the last group are developers so anyone that's willing to share her or his code the network may be to promote it may be to monetize it with a new model for example by getting Inc

refinements every time they caught these names used when we had all these groups connected together creating one common market that's the idea vision of : network that we're trying to build so where are we now we're on maintenance for more than a year actually we're having reworking use cases right now rendering is on minute and two additional use cases around the test net so scientific computation and the really

powerful one G wasn't so webassembly computation and how can you try it out it's really simple just go to this site that you can see here and install it and test it convection for pre-operational system for Windows and Mac OS and Ubuntu user interface if you want just to set your options as a provider and we had a common client interface if you want to try something for advanced and more complicated if you want to automate step on fuel to to test out this new use case that you spoke to us we have also additional product that is

called column on ingot and this one is for a powerful providers so if you have more than one machine if you can for example a company visited computers or if you have computers in the lab and one thing is no going on all of them just choose this solution and you can use it separately just to distribute the computation in your internal network outside the market or you can connect it to the column network via column only needed happen then pick one that you see in color picture and then your whole network of computers is represented as one back before now store development link this one you can also find on the

sides to patient so the next question how can one integrate with go on because those two where for request owns and providers so that top layer of this triangle that I was showing what if you're a developer and you want to add your code we go on and answer is use webassembly write your code in web assembly and just just put it in the column network like thing in your space that is live for more than a month allows you to do it so you just check out the documentation how

to do it exactly it's quite soon the good news about what personally is that you can compile the code which is written in other languages so if you have a code that is written in C C++ or rest you can just combine with each to what wasn't be binary and use it before they happen in the same way of using normal represent occult so what should you do in case that you don't want to write anything in any of these languages or if you want to do something more complicated then you have to wait a little bit longer for our tasks API so the next release that is going to

called clay does ppi is under development we can expect the first patient discipline and then like the amount of possibilities that will be open after reduction significantly increase so how this past KPI is going to look like what functional it is can you explain here you can just see a short preview it's going to be really simple so basically the app that you're going to create should be able to answer this simple and PC call so you asked to create a new task in your application

then keep going the next sub task for computation if there's provider we need to compute it then to verify the results and some big benchmarks to estimate help release your machine company to other machines in the network just like that as I mentioned keep track Adam there should be some updates about the speed yeah here is my favorite question so how you cannot verify my computation because there is this verified function that you've seen in Cascadia and there's no one single single answer because it can

get quite complicated for example for our first use case that is hundra who have this pretty advanced education to go Hakeem so actually request stories rendering cacao is smart small parts of the image then compare it with the result that he gets mu the provider and machine learning algorithm that is giving him a verdict because in rendering there is a high risk of some ended anism so there may be some differences in the pixels that's why it's so complicated fortunately it won't be like that under these cases with what wasn't in we're in a pretty good place because it's

relatively easy to make what is it because the Denise check our documentation there I saw some edge cases when you thought may not be deterministic especially right remember some problems with hands but not understand that architectures so what we're doing for what was simply we are introducing some level of redundancy so more than one provider is computing the same subtask be comparing internals and the same it looks simple but of course underneath there is like mathematical model and specialized algorithm you want to know the details please check our blog post about hidden does investigation how much does it cost to us :

unfortunately I won't give you any specific numbers because the : factor is not the one that is simply the price the market should be self-regulating that means that the requester who decides who how much less you wants to pay for the dust which some use cases there is a big price if you're able as a requester to estimate how long your presentation should take then the fixed price is the right model for you so you're just setting that's the time of my test and I'm willing to pay that much for the results until get the results as soon as possible that's a good model for a cluster but on if he's actually able to

estimate it so that requires some heuristic knowledge about your code and about estimation of what kind of machine do you need to actually execute the code in this time if you're not able to guess it then the solution is our pea pseudo model where you're paying for an hour of computation of course there's a steel is to provide our main GTO and take the Claremore hour's conversation detector value us down but from that also there's a mathematical model underneath it we have a series of blocks they still didn't check the exact equations but this drilling should be

better in case you not able to guess my second trial hit questions how are you going to protect my data and this question is relatively easy for providers so you want just whistle either the code that is me executed on your machine counter from the rest of your machine so that's pretty classical problems and the only natural answer is sandboxing so like some difficult some bots individual machines docker containers under continuous but assemblies unboxing we can discuss the

level of security that we're having here but that's pretty that are pretty select solutions the more interesting question is how are you going to protect the data that you sign it as a request so how are you going to protect protective coat how are you going to do that if you're sending your data to the cloud you not get needed protection you just tried you just trust the cloud provider and that's something that we can introduce in column we can add some form of distributed identity you still get nice nice ink so we can just complete your coat on the Aleutians that you're trusting what I believe that we can do

better the answer to this question is just that execution environment so in place these technologies it's quite early scientists it's being developed but that's the way how we can get her to a solution before hand tears the privacy about health that can also be the answer for the verification so better solution from the verification because this way you're sure that your cult was actually executed and that the code that you actually cent was executed in the money not modified way we're trying to work on a

solution called graphene that it's going to make the usage of trusted execution meant easier and we really really hope that that's going to be an asset for our network but not only phone only that's something that's gonna become a standard with the cloud computing what external components are you using it which I in house that's the question that we got from our community because calling is quite complicated there are many layers in performance for a long time we were doing many of them in-house so most of the gold in the world even conscription

but first not all of it as I mentioned at taking we talk the idiom was this missing market distributed processes come from and that we oppose using so you're using it even for payments using signal for deposits additional ratification to exist we want to use it even more in the future for licensing models for this distributed identity night was a big loss we used to have our own peer to peer networking connotation but that's not the way to go any further if you want make sure that the go on is

going to become a standard we have to conduct networking protocol so we are actively working on switching no Nikita P because Nick tucupi made enormous progress last year so immigration to go on from our internal p2p to network as we call it to me to Pinterest implementation is also for me quite soon and based on that we're going to love this resourceful layer because it's to our internal implementation right now and we want to act like protocol first unboxing as I mentioned we use the

standard solutions here so the external implementation and for this committee Education data protection we're working with our subcontractor invisible Phoenix lab and with Intel to work on graphene contact that is going to make this use it in distributed transept execution in wrangles ISM and the most important question why should I just go on you should just call them because that month and of a knowledge that we can't introduce so everyone get a freedom of

choice and a voice to specify what are your requirements about competition because right now if you want to rent a computing space you have an offer that you get from unit of the center the provider has involved but a requester or me very big company if you're a single customer you don't have tools to specify what exactly do you need and if you don't have this tool then the person that is a pairing you definitely power mine you can all your needs it's not important to them if you have ability to say what you need you can say that you

need for example privacy so something that won't be important in a big corporation and if you create a market that's going to be something plated you get again this freedom to set the price and set your opponents and the last buts really important question one and the game looks like how the world is going to look like in the children how do we want to imagine the future of go on usually in science fiction we've got this dark ant in top condition when you get corporations when he got syndicate

taking over the world that are holding all your data that are that they know everything about you if we don't want this sutra become a reality we need space for privacy reflection on each truths that can give us the freedom and now is to say what we believe if we need the price that's good if we need privacy say that we need privacy and what we hope the chicken goal is to build restore the network that can connect all the truth this work and we're goin back Tony maybe may just be one of the players but not one that

is holding everything exciting about everything and I really hope that you will care distributed products you're here because you're building this awesome solution as well we might need don't need assertion state now but we will need them the near studio sure I hope that together we will be able not to have this tank and two together but just have the truths that we thank you and I would like to thank all the reddit users that I mentioned here that ask us some questions and help them to prepare

for this presentation and if you have any questions do some boxing do still create some like attack vectors from provider like your share RAM or something case so that they still can access data now I mean on on the side of computer power provider so they still can access some of the data through RAM or cache sharing something for this depends on the level of sandboxing so we're having a different balance we have view on

machines that should and what do you think about comma morphic encryption for this purposes well I think that the question that you should ask to look at that is sitting over there he's the head of our research and he was like actually making some research about that on the government image thank you right now you the rent for rendering engines your only support blender right is there any rollout plan for like v-ray or maybe on the render engines we are also researching right now is stupid licensing model because all these

commercial services they can connect but they need good market incentives and we have this proposal where you can just create a short-term licensing on a blockchain that allows you to share your commercial license for a short period of time for a lot of users like working with lawyers you can see to create this proposal we will address small commercial solutions good and right now you only address CPU power or you also address GPU power and GPU however it's like Nvidia specific

question the techniques used for the payments move so like you're paying for the computation also for this additional services like I mentioned additional verification Travis cousin the shit's fine I'll go oh I also really really soon and you pay for this service with mr. comes form