Hashing It Out
Hashing It Out

Episode 70 · 1 year ago

Hashing It Out #70-Akash Network-Greg Osuri

ABOUT THIS EPISODE

Akash is a cloud for serverless computing. Using the incentivization schemes enabled by the Cosmos Network, Akash empowers architects and builders of the internet with the ability to utilize un-utilized computing resources. Conversely, those with excess computing power can open their systems for auction, enabling them to recuperate value from those resources. Corey and I have the pleasure of speaking with CEO Greg Osuri about their engineering effort and how they built this service.

LINKS

Injry welcome to hashing it out POCASTFOR. Wetalk to the tech, innovators bind blocked in introstructure anddecentralized netords. We dive into the weeds to get at Wyan how peole builedthis technology. The problems they face along the way come, listen and learnfrom the best in the business. You can join our wreck all right, we'll come back everybody tohash it out as always e an your host Gart Corpetti with my trestico host,that's always Callin Cuchet, say what's uper for Botar going it up, everybodycalling and today's episode we're talking to a caus dot network or acaution at work, and we have on the CEO, Greg Asury Greg. The normal thing tellus where you come from Ha Ganado space and then we'll just start talking abouta caush network. Sure I H, I hi everybody. I Hash Cory, Callingratviar! I am a a plot ECONOMII, a a computerscientist and a visual story to her began. My computer career UHprogramming H, since I was twelve o about twenty five years. I canallremember the time I did our program and Um the UM origins of ACAGA. Lee goes backall the way, uh to to t e time with a kid and and forme o growing up on a Formir in Indiaaffording a computer. I don't now how a program without usinga computer ou know using books and to get access to a computer. It's kind ofa big deal so um, so fast ballword. What we're doing tthat COEGE is making computing more more available to more people, so I think the journey really beg Havthere, but h, but toge more specifics. Really. Ah My background, I um moved to San Francisco Bor forteenyears. I began my career with building Dalasoner, cor, PRIBM and and H for big calicoes and big farmers and h. You know decided to move to the thehigh growing Hi High High Pacer Environment in Sevalley about thirtenyears ago and tarted a company called Angel Hag mingehig was a hack upon wits eccelerator. The idea was to bring topeop bring people together in a room. Um basically bring people together, thatr,not very, very exwore, I very outgoing and put them in a room. Partiticichours give them all the food and all the help you need and h and see what comes out of it. So wekind of made this idea of a hack, UPN, more mainstream, and by the time I leftthe company we we had about one fifty thousand deveeperson, VEICO system and in in about fifty cities around the wall and the the best thing that I learnedwas to understand, developpesented to rest and had n incredible opportunitythere to launch uh. Several companies from Hih Rauns and the biggest onewould be firebase few genal fibebas is firewas is the googles cloud Wer, and...

...it is what I consider to be one of themost loved devapertuling Um. You Know Panof, U top tothing, you know Wid,clearly be out there and to Lo stuff like that. That really hol people to get things done in the least amountof time is what I had experience launching andUm, and I stumble in to upon what I reallyrealize was big challenges after you develope your code. What happens then?What happens from the time you have your cod on? You get help too. The timeit goes to your users and turns out at that very piece which is calleddeployment, er extremely challenging, and so this was around twenty thirteen. Idecided to focus on that part of the development of Software and stumbleupon the technology called lunix containers Um very, very early in itspase Um and h found a lot of promise when it comes tosolving the deployment problem and startef focusing on it and got myselfinvolved h into projects that just started backthere and Docoris is a big one and Cubanetis um and and and really just you know, I was I guess I haply been the righttime- the right place and an in Americ guy called Jobera who created communities and Oncei Yo, knowsatis dam. All I crosec right before the laugh. I got. I decided gettingwall, much deeper, therewas, a lot of promise and and and that's how Istarted working more seriously in the crowd,auimatianl space and Todayco en it. Leas is used by eighty percent of thecloud. So that's where really how my journay began with making soffer muchsimpler to to deploy Um? And? U We, we started a castionet wr, the company, you know the kindesaracasianer comes fromoverclock labs. Our mission was to really take Commoditas compute to alevel that it can be deployed at the edge, an hy hyperformance of SocialOrledency, and what we discovered was there is a tonoc capacity sitting indata centers about eighty five percent of the capacity sitting. In four point,might eight million data centers is not used and and and surprisingly, there's this-this new new sort of like industry, that was just growing rapidly calledthe cloud service providers primary led Bam, an Ogaac Sop, uh, that's gainingmarket share in in capitalizing or arbitraging aces for Likeo, better word.So so there's this incredible capacity, that's not being used, and there istheres a few companies that are capitalizing on knowledge and a and Wefelt that it wasfundamentally broken. And so we started experimenting wit, Unlockin, thecapacity by by creating marketplace and and wewhenante designs, marketmartplace. We didn't want to design it in a way that is, is going to die after if get stop working on aproject. So we come from a big open source background and we wanted todesign the softare in the matter. That is pure peer H, so that looks amme likeget you know if you get get notr, essentially Um, not Centaliza, and they have this likepoolar push model and we wanted to do something like get with Tustabl efeatures and THAs hobbies. We endd up boing blocdchain, so the the wholehistory really comes back to Wer Um is that we hade this massene newscapacity, there's an opportunity from a marketplace then there and when wedecided to do a marketplace, if we want...

...to do it in a decent, lave manner andappear of o Manial matter, Gitwas major inspiration and if you hadtrusibito gate block chain happens, so that's real journey of ofogowork. Firstquestion on that one um: You have this massive UN news capacityand I come from an HPC background at even like scientific clusters have agood portion of Um their their computation doubt in terms of likeoverletalization and I'd. Imagine us the massive data senters, it's evenmore so because they're a lot more commoditized in terms of the harborbeing used. What what's the problem wha w? Why is it under utilized? Is it lackof access? Is it affordability? Is it there's not enough people who need torun jobs and people built datisinters that are too large? What is it so? I think it comes down to TWOCOproblems. The first problem is Um, a Deplie prepeique. So if you are acomputer provider, say you're Walmart or say you are turbo tax example right,turb ttacks during their tax season, their utiizationssomewhere aroundninety seven percent in the real lembers are and when they are h doingthe non taxes in just nine months of the year, they they're oulization likethree percent, but you still need that to passte. Athree must be ear right, and so the peak planning is te numberone caused for a newscapacity and the second, because whon newscapacty issome optimal distribution of workclothes. So Um, if you hear theterm, you know sequel servers, Ornonoma Serviar abservers what they really means: The they're dedicating this entiremachine for one function. So they usually run these machines, ot e, veryhomogeneous matter in homogians architectures, so that causesignificant. You know inefitiop en come distoutionworkcodes. So if be Gan, H, somehow really capitalize on onyou capacity andsomehow capitalies on distribution of jobs or workloates than scheduling,then in a manner that that's driven by a performance envelope, you can solethe efficiency problems so yeah. It's IT'S B planning and inefficient useg orinish efficient, architectures wer for scheduling, so your you're, I guess audience or or I guess desired. Peoplewho to provide this computer resource is not going to be the big box. Peoplewho already have cloud services, hat's going to be 'cause. They havecontinerazation and kind of m they cost out evening or whatdrresof white paper you kind of figure out how they cost out and price theirservices. It's people who who have planned for this? U Certain amount ofHardwar intrastructure and they have it, but don't utilize it, but a certainpart of the year so that they can provide the same resources for a muchsmaller amount. Is that is that about right or there is it broader than that. So, if you're asking, who providesyou're dona cash nutwer that it's a wid wide wide range right, so all the way from home, depoytments Um uh to raspberry pies to you, know Megad Addi, centers,doyouncobrolitors and Um. You know it it! It's IT'S INCREDIBLE! Some of the TATwo provided that we've been talking to H, still have massive onder utralsproblems every time, there's a physical box somewhere. That...

...is only your life, no matter what right.So it's really a wide wite range of people that are using so be biggestproviders. So far we saw on that on the test: Net machile learng companies, hthey're, deployed hardware in their facilities on the site, and thesecompanies use them, maybe one or two inow times o year times times a day,maybe vrs and and most of the time they're not using it, and they want toyou know recapitalize or recouse of the capital, because so his Hardar is verybey expensive. Two, you know I have superman rightright behind me, which is my home computer, which I barely use, butthat's Goin to be onon an hour. So who do you see the end user? Being,though, I think that's what cor might have been getting at Nowthe Day likeIail? Oh really, okay, sorry, I mst miscender stood in I a misunderstood. Iapologise, but who do you see consuming these resources? What do you see their user experiencebeing so? What kind of availability guarantees do they have? You know there's? Obviously, no. I tendto see like theres one one o one service level. Ou Know Agreemet GRSO. How do you who do you see the target audience forusing this unused? Compute power? Is You know right? So our biggest demand comescomes from machine learning and data intends to workoude medical, becausethe cast for these to run these jobs areexponentially. Increasing 'cause the size of danae processing is is growing,especially if you're deep, leading opition in company, sosome of the biggest users today h spend about twelve million dollars of a cloudand some for smaller size users spend about five hundred thousand dollarsfrom the CLOCB, so these are large scale, garring, tensef application andand and run by growth stage companies that are extremely sensitive to cost.So any cars advantage we provide them is extremely attractive using a cautious model. The costadvantage you know is is about four to five times lower than the current cloud,so that's an extremely attractive centerfor permassing your companies to to use and from a user experience standpointUm. I believe a cash command line isperhaps the best command line experience when it comes deployingperiod, not cloud or not. Blatching, and the reason for that really, youknow, look at my background nd most of my work. I did any library that I have over that haveover tousand stars were about making Um. You know H, givit up experience, ONUESISTEMC ore,delightful, so so my background wis making commanlineinterpaces and designing n developing interpaces when combined with wwithcrow deployment. We ended up creating an incredibly USAL tool and I believe one of her partners h putting a survey about phat about ahundred depeor person when they ask some what tool block chain crypto do,you think, has a most debest developer experience and think Tory penty peoplesaid Ar cash so that that's really testament that's Reay A to toto toexperience and we do have a wall of live people. ARCASTOR network, slashfor,love, you'll, see p what people talk about their experiences. Wan Come Stildeploying on on on a cash, so it's extremely productive interfaces andvery delightful interpiaces. We design from a SLA guarantee standpoint. This is a decent life clous, so thereare no guarantees right, er nor...

...contracts, and although we do have mechanisms fortopfall tolerance and really the design for the for the systems goes back todesigning for hyperscale when you desine for hyperscale, you do notassume trust or you do not has zume performsat the age. So you design systems in a wave ther are that th they're highy versitile in Highry,highly uh, incorperable and flexible. So we design the continerization tocollege. That we've been working on is design in a manner that you can switchover um pro machine to machine in case of a nolfailure in a manner that the spall tol Ram are also in a manner that you donot lose a session right. So else, as someone experiencing a faull tolerantevent will have no idea that you know, there's a nose whitch over while usinga cash. So so so you know, catually the hyperscalearchitecture designing away veri similar to have googl most of thehyperscire architectures are depoyed Um in a way not to disrupt the actualanduser experience, while doing a fault toller enevent. Ontop of that, I'm curious M for people who have these larger jobs that requirea lot of resources. Um Are Those d T. does that mean theyrequire Um, specific subsets of the networkthat are all basically located. I hi physical location, like in all these NSntto, an those types of things be distributed across Um, whatever devices that are capable ofrunning things and a so had he had is that like network lanes yo come in toplay 'cause I my background's in in scientific computing. So it's most ofthe Times you had to use casters that had very fast interconnects, but I'm carious I', like there's no way you could provide thattype of service in a ditrubuter network, so the jobs are going to be doing. arethings like machine learning, but even though sometimes based on the size ofthe job need to be in the same physical ocation? Is there a way you can handlethat? I know that's what like Golam had troubles with in the past in terms ofthis kind of Blotshin, compute, bottel right H, so local resaurancy versus remote eensyright. So the the Um it really comes down turns out whenwe did when we, you know we're doing a lot more research yea with the user S,it really comes on toport price performance. So when you have a priceperformance, metric uh, where the cost is insignificant, Lagency becomesinsignificant as well, and if you really categorize a kind of wor clothesin a betyway highl here, two types of workoes laden seasons to work close andthere's batch optemal workclubs right. So when you do batch optimnality, u residency, even though it it looks like you do. You do requirea Webe. The requirement really is diin the cost so um all tesis is the cost isExponatura over what like eight times, nine tons lover. A agency becomes lessimportant for BATCHOPMO rortos, a Livi becomes extremely important when youhave you know long running jobs, which isessentially that need to respond to a user bend with a hundred miler secondsMHM. Now what o CAS provides is in extremely flexible deploymnt arcitectuatingchoose the Lord that you want to deploy it right. So when you have a job thatrequires local residenc local clusters, you can choose to deploy in a clusterthat meets x amount of nor capacity...

...right and only then to pay the job onthe plus, if not don't so. What I cash does really is open up the market tothis. These MASSIV plusters two small little deployments all over the worldand gives the sobereignty to the user to choose what they want. Of course. U Th the multi, color archiecture INA CASalso means the cloud companies plug into CAS RETL, which they're doingright. Now we have SOMTH, too provider sliking Ino to a Cox that have a lot oflocal coster availability. So one of case provets is is extremelyflexible moral, where you kids want to choose a local resident, trustherwithwhere, you know minimum Lord a requirement you can choose to do thatand once thet job is complete, you can choose to sitch over to a more cost: optimal deplarment stuctures, soUm, it's not limiting, but rather aditiveon top of the existing in the structure. Another. If you look at a Castiomr inother Lans H, you can look at it as the multi o Lon Pacfrom, as well using thebig clubs right. So I it's really. A technology that you know is is designed to move work clotes in in a proponence in a cost. Envlopedefine te user H and we provide is extremely H. Simplto use u contract language called SDL or sackdefinition, language that lets se orchestrate work, clothes, UM UH! Youknow simple cloud functions to ve advance, deep learning, e Multiregiol,Multpizomoa custres all over the world in one single Pale. I G. It seems as though the way youbuilt it um gets around or fixes some of the issuesthat I think that earlier distributed, blotkind compute platforms, ere havingand that you had like how do you prove a job was performed appropriately,you're, not you're, not necessarily creating a marketplace for jobs whereyou set up a job and get an answer back you're ou're, you're, CIIN omarke placefor the resources themselves, thanconmen so and into containers. Iguess associated with it right, so verification Um in compute isan extremely hard problem, especially when you're doing general curposecomputer. It sovucation is possible when you have access to the language Hor o the framework you can plug in, but when you're doing general purposeconpute, it's it's mext impossible, almostimpractical. So, instead of taking aveification model, we took a Weber cruss model like today. When youdeploy a workclode on amazonor or or Google you just. You know, assumed thattheyare telling the truth and they're going to ask. I there's no guaranteethat they, you know theyg Providet, the workcords that they said they're goingto provide. So we emulated that that kind of pattern, but wewen once keptfurther on making sure that we can create a systemh that is flexible in terms of I you know, I'l be trading, trust it. So a web ofcrossmorol was was a lot more octimal than avetcash morol and H, and when you have dishonest notes, ththe reputation of a dishonest nowd should give information um enough tothe user to make a decision whether deploy on that dishoest, nor an art and h. So we don't sapply. Thepeporical does not limit you in terms of Uh Youre choosing and what now you want toemploy base on honesty, a dishonest, it'sreally like pushing sovereignty to a user and let the use it EI Kan. Do...

No ou had a few questions in terms oflike Um a I noticed. Well, I did notice on youryourub. You know landing pace of your website. You listed stx under your stack in one ofthe graphics. You got a curious if you're, actually using US cx yet or ifit's something on in your pipeline and f. So what does that look like yeah? So we are using. I yet's we, thisideol from one of the partners that you know amusing them over, where theywould deplay something on Amazon and, and that doesn't have, is jecks anddump your keys and literally as a a noisy, neighbor problem. Ye Call itright and depliyd with th Nexcek one time. On top of that completely youknow giving Yo th so amazing on claim, that's impossible, so turns Ou, ams anddoesn't proes CEKS. Yet I mean I don't know if they're going to Um and h yeah,so it works, it works in a UH. It does require a certain level ofexperties from a user standpoint, to understand how the use is checks, but it works the only limitation orlike it's a trade off between security and performance. So when you have like hyper performing systems, if you'regoing to incrp all your you know everything in it, especially you runtime's going to perform very slowly and you ded understand those trade offs asa developer, to use thes Cek feature and U Sinc. ACASH is a divers marketplace with any type ofcompute Um, so sjeks or no trust on or you know, theprocessor eqall and secure on clotesis really dependent on the on the computer and h. If you go to a cashright now, if you know chitial providers, you you'l know what providesSamei CAS and what proers don't so yeah? It's? U It's! It's jx! It's a primitiveONACACH, yeah wat! I sa I'm actually curious, is like ticking through whatyour kind of doing there in the sore say. Have you released that that yet isthat? Did I miss it where? Where where could I take a look at some of that, it's Yeh it's available right now youcan go use it its ceks, but you got to bring you own run time. So, up to you,oh I see you can literally. I gott you, okay, so you can literally okay got itokay, so I somebody brings hour edesttx one time you can deploy it using me. ACO network and the individualoss themselves will lendtheir s so that they actually do the signing rightlike that securely for you.Okay, I got it cool, so rentime is very I mean you have fewruntimes you. Werecommend UNJNO on time, they're amazing company as what they do. Ibelieve some of her friends are working on their own run times. Um Mobile coinis working on her own time. Last I heard h there are lots of lots of runtimes, thatare being developed. They have different, you know agastation anddifferent mechanisms and ows jeks Um. We haven't seen m any major open source and Pomantatonfor stx. Yet so I think the technology is like still early and under development, but that's I'm pretty certain. That'sgoing ta change like the next year. I think so too. So yeah, that's reallyinteresting and so Um. What what kind of incentivization model doesa CA network have like you, like overclock labs, have for putting ustogether like whow? Are you tied to this network? Are you minting coinslike how? How are you guys actually making money off of this? How do wemake money? No BILLIO lot a question: how does any open Os Copay make moneyUm? We we don't make money from a castion work per se H, but we makemoney if our curson cork succeeds. So...

...our current tomorrow is to sellhardware to support the SECO system and weare doing that using superminey and second business model is to providemanrage services for ladge providers that orlarge customers that want 't employwithout couching the Camal land. We we do it for them and we created ouvr o'clock lass. When we began an thecompany or our CIESES was we want to unlock theedge by creating a software that makes age deployment as easy, as you know,asingle diploma and Herico right and U and UH. So we Y and we needed thisnetwork in order to succeed h with our the original business. Now now the thatworkas is depoyed and and as an apor grows H, we expect to see a lot more usage services, especially IAING fromamaer. So at the same point OIT's Obe interestingbelieve. It's a veryinteresting model that that, in that a lot of erar experimenting, whene,it comes opens a sofcere. So so far, the business models whoopee of softwarehave been primarily subscription, driven and open core driven MHM. Butnow we're experiencing the Third Business Morel, which is tokenize,opens a software. We call it therhe. Has this incredible incentive air thetoken layer on top of the open source h, you know platform where the the Gara as well, Imean Sosorcco as well as AAI Osor. So how do you monitize? A lot of us are-are still trying to figure that out ohehonest with N and our our solutionis stetting, Hardwar yeah, that's definitely Um. I don't think anyant tofigure that out. Yet I tyet, I think, there's quite a bit of potential therecoming from working for a company trying to do something similar in termsof a a token open source project think nows a great time to even getinto the economics of the network, because how you lay out the incentives and the key players of the network andhow that token flows across that so whith. The life cycle of the tokens areM. we'll say a lot about your potential tosucceed, because of that iequilybium is inappropriate, that it can never reallyscale to baby the performance. o Ther never could be right right right, so RTH CIS, this whole decentralized and whol tokeneconomy really began with a paper a published calledwoot strapping, a premarket by borrowing from the future. Sosenshematisis was inat two side on the marketplace th. The first challenge the markethouse is all is the demand O plapardax. What comes corst right and UH? The market has to Sollthis problem inorder to creatate clibrium h between demand supplies so that the thethematican can can unlaw the network facts which is which is what drives ago,O the market and the question ofis. How do we? How do you do that? So our taseswas when you boots up supply to a pointthat it's extremely attractive to e, to the demand, let's say: Cost,for example, Um the the demand side, Wi will catch up so and we can o nowreduce a cost by subsidizing supply to recruit a model where Um? U, as a supplier, the that someone that Hahave compeed, essentially hase Zour...

...risk to to dedicate the comput to netBerg and the risk is midicated after a subsidy, the network prolikes so and uh h, an and incendiwizes the providers toreduce cost keeps keep costs as Lell as possible right so um really you'reboring from the future to bootstap the the present and really what you'redoing es you're, creating the effects of liquility on on on a agreeable price point what theinflation's going to be like. So the paper arow really addresses how do we?How do we create inflation in a manner that is acceptable for a price point?What is the pricebor look like right and one of the things we did when itcomes to Proofo? I mean the system uses a poof or sakemechanism, incomes, consensus and the token her used to provide a Conomi,Securi thebronc chain and t was interesting w with this talking, um comes o Po stake systems or whe incomes.INFLATIONA systems, as the the rewards are essentially dependent or driven by the MODOLstakebule, as well as the time ustake right. So so that creates a much better price performance when itcomes to to talk in price or in in Wulmart in a bare market, we expectpeople to lock up tokens for a much shorter period, wher in a bull market,to expect people to lock up tokens for a longer period, which translates tothe amount of invlation rate, so in a bull market infration can be higher and and because there's less price pressure, that's going to be kbut in a bear.Marketment, you have a lot of price pressure, the token lock cops or theinflations. I bean a lot lower, so the the Toke models designed to be adaptevewhen it comes to Um when it comes to performance from amarket standpoint and from a token capitalization flaw,structure whe when the capital comes in an CAPL wors out, the capital comessent really from the from the user using netword and thatgets distributed to the providers. Andto the stakers and a whole lot of state corders in the in the ecosystemand h the Um in it, and I believe the the the the inflation rewards aredesigned to. First of all, creating centives porpravized woul, alsocretincentives for Tor the stakers to provide conmomy securias change like soum. That's a Vey, very high level. A andthe token is, is benot important ther. The token is really uh functions on thebackground. It doesn't mandate you to use the token ounodeto use. The networtand a car supports a multicurrency settlement, megalismwhere you can steatle using that coiner any o USTC. We assume a lot of laerfolks are yeah yeah, you know getting to the stable, stillcoin, russtabecoring itization. So of course I a wide rangeof token in spocerement and of course there are a lot ofbenefits upo using that Asho giverent comes to cost and and other things, butbut H, but providing a multicurrenccy settlement makers and well believesolves quite a bit of problems is o Wy, very opers, because the carsn at Wor,you know being a cloud infrastructure is. Is You know, as you know, in a cloudsystem, the solution comes from it from Barios different systems workingtogether right. So as foror example,...

...you have new Cycher, which does keymanagement, more storgay, which does arcaevo storage or helium, which doesyou know long range, Byfi, networking and the towns of incredible projectright now in in in in the in the CLOUDICO system, to perform criticalfunctions so that you require, as a solution, developer t o to interpratewith so so with a car. She can literally payusing any token and use all these services Witho ar having to hold theirtokens Um and have that level of flexibility. An that level of comfort H,which we believe is super important for podriving, an option for for decentplas CA, which is which is a huge problem to hem, because you cannotincooporate or you cant, Automate Sin Ditan Matter. How do you pay price aspecific unit of Compute, especially when? And does the? I guess, variationof price of the token itself Com into play with that? So it's a very challen, you probem, wedon't trice the COMPUTR alki choses a rivors action, a model where theproviders and and the payers the tenants h price the compute usingauction, because you know every attempt to price computehas ende en uterfailure. U S being, I think a big example on how the t priceam or commoditys ramevery time you comotidie something turns out. You know it's a classical game,PHETICAL PROBLEM RIGHT! You have the you know the NASTHY Culibrin problem.It call it Ingann theory. Every time, then you create a commodity. There'salways going to be room for people to Cheat Right, provide you uh this sake: comod Tha, that's oflesser quality, so computers very, very hard to com, ainize instea of trying tohomoritize as GR free, Martagos Revin, my auction that that's been or teses,and because how many vable Hoa types of Rad do you have or how many types oflike clockspeed for CPUs Te be variable right, there's a lot of verybility whenit comes to what kind of deviss you're gon to be using. So it'slike diamonds and it's not like gold, AGREC GI. How do you prove the commodity? That's being supplied is equal to loo. It'slike. I know that, proving that a computer executed a program uh you know to you know in good faith issomething you can't do, but I im not really sure. Did you speak to thisabout? How do you domake any attempts to make sure that the person who isputting their resources up for auction or actually have those resourcesavailable e they oa or GPS? And you verify that s? That's actually the casein how do you do that? So there are answer? U Question. You can'tinsarpactial right. There are mechanisms you can deploy to prove, butif you have to, if they user have to prove every time they provide somethingthat they tell. The truth is just not going to happen so uh. You know the best pausible proofthat you have is like an asymmetric memory Hart function. If you want totes memory right, it'ist like an encryption hack toessentially a wrong a job so that th that requires exol Mo memormeats or not so what we do is we provide a instead of proving weprovided trustbes modow right. So if I say that this provider is legit and I'msomeone with high reputation by association, thisCowa has an operation like just like PGP, essentially so um, and if thispliter is lying somehow and I ou know- run a Chileng down the Toideran, theproliter know reproved upohit is lying.

I take my trasspoints out and morepeople are more oer people running the challenge on T. Qaes are continuouslydistrusting. That means Ou, as users should probably distruss the pier re, areputation style mechanism, Buti similar to help profof stake works. Youknow, there's no incentives involved in terms of like proving that someonedoesn't have a specific resource. Doesn't win net anybody any money? Can you prove it? You can't even proofof the adults. So here's the problem I', I'm Seying, I am a provider. Ok, Iprovide a particular type of resource Su o say I happen to have a buttload of SX enabled systems forsome reason, and I know that there's a sudden spike in Deman forasgx, forwhatever reason: okay, cool hat's tikeit on this ce graphic sores, Idon't care Um and and like I notice, a a competitor POPs up. So what do I do?I go ahead and Smith. Some under my other accounts, my you know my sake,account Um and I submit a bunch of requests and then I say: No, you didn'tdo it right, Ding in your trust, nood! You didn't. Do it right digging yourtrust, Ou and then I knock out my competitor Um, what stopping this kind of snario fromhappening. So that's why reputations so important. So if Youwho are a nod thathas vailow reputation and you're, not nothing all you compare uh, then you know you're you're, knockingout, doesn't really matter right. But if youare someone with high reputation,that's knocking out. That means there's some Weig to it, but tisorry tointerrupt it. I think I'm just there's a district here 'cause, I'm notunderstanding what the word node means a I'm very sorry to interrupt, but ththe. When you said note. I think I think you know a compute note. I thinksomebody as a provider. So I think if there's there's two there's providersthat there's consumers. You know the producers Anconsimer just peopleconsuming comput power and this p people rroducing coupe power. Thepeople produce it Coule, be a power need to trust its mechanism to ratethat they are actually giving providing the crrect bility in the rrep power.That memory and all the stuff t they need actually, like P G, take care ofthe job. All you care about is the job gets done right and it gets naing goodfaith. Then there's the consumer side Um, but I don't see a trust bottlenecessary for them and there's no way to actually verify that they're doingthe trust bottle correctly. You see what I'm saying so like they don't. Soif, if I may produce her, I produce I may I ay providher, I I provide M. Iprovide my compute power and I see competitor who's, also providingcompute power. I could have hey. I could easily have a myriad of mconsumer who have actually used use to the service F, every so often justdumping, my money back into myself, Kindin a thing or use other servicesjust for whatever reason they have a small reputation or a history on themand then theyd up a job, onomy competitor and say no, you didn'! No,you didn't do the job right. This is wrong. What do you? How do you rizekind of things yeah? So we're talking about a civil problem right, so you're?Essentially, every time you deploy something there' a cost. There is a a cost that goes directly tothe provider and the portion a cake b. You call it that goest, Toboiintes andothing about twenty percent take p right now, so you cast of just sort oflike reducing your uh competators reputation getingcreasingly higher the more number of jobs you do so essentially, there is n today in t e on the cloudday right, so you have Amazon Gogamaxa. So three companies essentially Um prowinting compute, now yeah as a user.What I would do is, I would run a benchmark on Amazon and go I amicrosop right and based on the based on t e brenchmark. I established my ownreputation gauge and I'll, be like okay.

You know, I'mgone the same instance h.You know same configuration, amblance just performed better, I'm not going touse Ghool. Now. Imagine me taking my reputation gauge and putting thatonline, so other users can see my Daa Wat. What I experience rightnow, it's up to the user to make the decision whether they want to deply onGoogle or Microoft, our Google or Amazon, and run their own Brenchmarks,or just trust me that imtem the truth and go with mydara right. So that'sreally the mall that we took with a car. So there's no the reputation of we providing does not limit of scheduling to t he job, but itgives you enough information to the tenant, tha th, the deplying, the job,to make a decision themselves on whether they wanted to Puy a job or not.Like certain case of a Sibil, where provider you know a is trying todiscredit towier B and I'm a tenant, and I uh you knowit's up to me whether I want to take that ipetation and run withitor. It's up tome whether I I can run these jobs and have my own reputation to challenge their reculation given bySabo the provader aqreated. So that's really why I said we have a webcrust model its a trustway system. It's not a Um! You know permissionless vepiable system, that'sgoing to limit your job deplyinment based on some factor, so I see okayyeah. I ssaid that I thought you were scribing a trusta point system toparticularly nos but you're, actually just having people report feeback likeyelp correct, so like revus. Eventually I got a few questions from here. Thenyou start to kind of kind of start to rap up here. What theHell is, a is is supermany, and why do I need one so superminny is: are, is a homeappline that Um is essentially a medi super computer and h? It is, I think, as powerful asit Crae semine, which, which is the most the fastest supercomputer in twothousands s more powerful than that and it from if you're, if you're a user. Insimple terms, it makes you more money that you spent purchasing the devise inthe first year, so the ROI is expected to be somewhere around hundred and tenpercent the first year and Um and uh th. The cool thing was Tobe com supermefrom from a usecesstandpoint is Um. It gives you a mechanism to bring thecloud to your house right and it has an abser moral that lets. You deploy yourfavorite recent teas applications. Now we have eath to support, lie poof steakchains that you can run from the comfort of your couch at significantlylowercost Tand, what woul otherwise pay on the clouder deploytment today andsome of the quale applications. wher were we launching Supermani, wherethere's WI, Olkid thegepn, ou, node and sentnel another increbble Wecam see soyou can tun to be piano directly from Your House and totally skip the skip.The big you know, big cloud and and also another cool application.Thatwe sing subliminy is proviling an intence layer to this longine network. That helium has queeled Phitosope that don't know what Hiliamas Hilium is is is is beautiful? U, you...

...know, project that introduces a horsspot along range hot spot with about ten mile radios. Now I on sitting O my house, so what this houspot does is is creat,so mech network for low power, ioote devises to directly connect overvirelessly using this O, essentially to to Ospot and and uh and Supermany, when it's connected to ato helium, devise powards inflrence layer, because lowf devises are superlopowered and they designed to be Um um they desiny just like do the sensingand acting and they don't have the kind of intelligence baked in at the age todo any meaningful inference. So the infrence player is what yea supermannisolves for. So you imagine now you're, essentially creating aagent prostructure that skips the big tellcos and the big clouds alltogetherso um that siggnificantly reduces the thecost indepart cost envelope whene it comes o deploying these age, ittyvisers thet's, just one of one of the super call use cases. Po O be maning.So, depending on like what angle you wantto look at if you're a hardware hacker Um, you want a Supermini at home,because this is your Atstore essentially for your house and it'sextremely mardular. You can metally plag out plug in different boards thatthawould be releasing or if you're, some on Hee case for sovereignty, an inprivacy Um. You know, having supermines at home, gives you this thisopportunity to control your own Dera and have the dela where you want to deploy so Um. I don't know if you saw our dok, but wehave this. This deployment guide to deploy Matrix,chakserver MSO. Imagine having supermanes in your house in my houseand be both running matrix on it, and we can connect directly in a purpermanner and have wor soip fully secure communications into an encrypted. Thatskips the cloud alltogether right so th e are a lot of like a lot of likeinteresting usecases, that we are locking Um the supermany that focus ofsovereignty um both of a computer and Arra, and in doing so you're contributing. Areyou building the next next layer of the web tha the web three forthe for the edge right, so Um Yeah? That's a really re! Really WatkSemiy in the box. Ooks like I'e, actually really like to gave my handssome one of those. How do I do that? We are shipping fifteen of thesedevises in five weeks. H. If you want to get one of those devisors, you cango to acastat network, slash, Supermani, slash, reserve and put your name on it. The firstfifte wershipping is to do a fiel test, and in order to qualify for these these boxes, you need to havesome level ovlink skills, and so we we really our goal with the initial shipment isto get that feedback and h and UH and h. You know I work with you an seehow re going to function, the field and we're shipping about hundred of thesedevises in me, Um, so an in in more so in the mecoing month. So wedoing a phase released, Relese and UH yeah, uh, yeah, really g get in touchand O Shou be on. The next question is m.just kinhat's is something that I I'm just kind of curious to this point. Ihave a I I'm one of those weird people thatjust has random computing resources...

...sitting in their house, not doinganything. I have I a sixteen COR compute note waiting to do something, what's theprocess of who signing up and getting it on their acastion at work. And whatcan I expect from that process? Ifyu the process is extremely simpleand if you go to a do star, ASIO nework, there's a guide or how to Um become aprovider on en acrt and what our users tell fromthe tritlers. We go to acros an Ewik, slash, love you'll, see a lot of thefeedback on Rittor as to whatther experience. He is like deploying, andalmost all of them found deploying or unlocking our yourcompute, that you have a home Wav simper than you can deploy on the cod.So it's a really simple Um um we to do it and we kindaf created tons oftooling around h being a provider and operating a eficiency. So we have a a tool called disco. Thiscos, transper, decentalized infrastructure forServalis computing operations is essentially a frameword for providersto manage or notes at gives. You amazing, tooling, all the way fromcibinities to N, observe aaability, tooling, Grapana and prometeas and allkinds of tools you need to like, like like make in your your computer n andbe informed when, when things happen, Um- and in that process, you know youcan you can install the AAS for wider, not on Onthe, on your on your machineand and just you know, yeah just unlok it as as simple as sthat cre uh. First, I one thank you for Gom on Oshow Yower a recommendation from one of our Um Sands of the network akin fromSychrocor. He he said he'd like to hear from Ye all and recommend er Yolookingto it and I m mglad Yo didt yeah yeah, I love CC Cania be Metcan, I believe Mthrough the twitters and he's one ofthe proliters on Oak Nowork, and he did exactly what you wanted. You asked meearlier like how easy it is to get on. I A workan once he experienced the Softwarei was obviousthat this is omething Ho. He wants to spend his time on and I I believethat's how we got intuduce allright. Well thank COO, shous group,Congohadi kind of one more thing like has ther, already been like userstories that have really like come out of that. That are just compelling thatyou think might Ike somebody acsually use this system to do this coolscientific operation, for instance, like what? What kind of experiences arepeople having so far besides, you know lending their resources but actuallyconsuming the resources. Is there anything on that end? Yet? Oh yeah, yeah, yeah, consor them Um again likeak a nervc social Lov that really describes youhe experiences. I think myfavorite one would be Um Alexellist, who created open fast. An open fast is the most wilely adopted function as a service platform and his school story was he created this Um, this cloud load balancer. I think that's how you describe. It iscalled interlets what it does it exposive, an Ip address from your houseUm using a Lo, a lottle balancer sitting somewhere on the cloud and Um, and so then you can O can share your y.u know your service sitting inside a nat in Thi, nely house right, and hedid that using a carge uh and his...

...experience was like he was gowng away because now you knowall of a sudden he hase thi new usecase of just exposing your local computer toto to to the Goup right, and that was verly hard to do witholacash andbecause of the iguess, the permissions nature ofit anan and an the extreme program birty thatpermissionless mechanism provides Mak so that he can bake this, like you know,decent lises, Lo balance from all like Tink, O Carit, now um into his courseoftraye and next thing. I know he's startinghe's just running all kinds of things, all kinds of like machine laning,infrences m. He sent me a picture that colorized my photograph from like areal black and white pictures, all kinds of cool things. People are doingand an the most exciting thing that that people are really excitedabout is the privacy protection privacy guarantees Um, runing, Matrix serviceright. So we we ran a challenge where we wanted people to Um to get their own matrix domain. If youra e, if you use riht or matrix when you sign up you're, usually they use thenameis like n Gregat, Jo Ohord. So instead of that get your own name right,Um and h create enough noves in Matrix, where you're communicating uh privately one way toe to another in ina pty par mannener. So that is like that's incredible. Ewe Got excit excitement from fromfromthe public for that kind of deployment right, so Um, there's a lotof people that are excited about acash are machine learning and and privacy Um.You know seeking a Fox awesome. I actually like the fact thatpeople are incentiviting her coming up. I new ways to build theround matrixservers on I like that project, a lot and and connected with them Tho to workthan idea. Yeah. Yes, because, like I don't knowif you know, but sixty percent of tering anoso running on on eight ovyears, right now, whichis kind of sad that that's one of the reasons why wewant to fix t e many, the big gold for disco is like well, we are um they're giving up. You know progress for convedience and, of course, Aus and Google Ar Betyconovenient to deploy your workclothes and we're giving a convenience, becausewe don't have ha capability, Otis, anise Mannar, one of the big reasonswhere bescarted Tisgo was to solve that problem, and you know in that journeywe ended up treating a node that you can deploy in your house,extremely convenient, extremely caust, effective, more so t an than the cloudin a hope that we can decentalize the supposedly decent classworld today. So, Y, U K, it doesn't really matter how decent it was inNetrork case really comes down to your physical deployment. If you're runningon a in O on Amazon Um that I don't consider that Disenliz at all,especially when the majority network is running on a single crown forwiter, Iwould I would igroe with that. The best part is no one is using itsCEK, so in caseone, as adversary wants to take over if you're in that word,they can just implet dump the keys that has physical axes. I'm not saying thatjust basis gn I do are, or R or ex or current administrationis going to haveany intention to shet down if theybut. If they intend to do so, it can be donenmarromax right. So you don't have he problem ina corge, because, even though you have physical axes, ssx a Gointa make surethat that that that's impossible...

...ers on so wher. Do people go to Youve,set it a few times, but just TA RECAP: The vere episode where Yoll go to learnmore and get a hold of your guess. Acada network is where you find uswe're very active in a chat: Rite Chat, figur, Arcasta nevworks, like Shat, butYo Wan to get started dog startacastn network is your best place and H, peolereally like Adog, and we work very hard at it,and we can, you know, ' getting started line. Ther uh, you know start using it.The best way to learn about Aukashi is to use it, and once you use thateverybody that use has their own description of what Akash is. That'salways exciting. Cormedi t to hear so insor me describing what it is. You use it and you tell me what do youthink it is outstanding Oll? I would start setting up my my random noutes today and, let me know your Matrix Service Wdo have it INSEDEWITES, that's not coming up by the way so um, so so in about two weeks. So, if you aresomeone that likes to operate an odor have aspirations to to you know: Gaine Om prosperity in thisnew new proofastak world, plice joiner in Saneves, deset and uh it it's a funexercise. It's I set of fall poor set of set of for challenges that walk youthrough how to set how to become provided and how to start using a causand O arvaiating whatnot and of course, insation means you get token. So you know it's it's a fun game. The lastchallenge: We we had the founder challenges. People have so much fun, Um at a p e the passes, grewing deenriftcloud with about forty five or soproviders, fully decent lives and h.We had about three eighty applications running actively: DEPOIT ONA cach rightnow we have coming close to that bad, but uh. A A lot of that came through likeincentivize aton programs, wo system to LAFA awesome, so te Tau X ot go on again yeah greatepisode, axalyeahdraing. so much Ti Isis on.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (108)