ABOUT THIS EPISODE
Episode 34 · 2 years ago
SHARE THIS EPISODE
Episode 34 · 2 years ago
Hashing It Out #34: Storj - Shawn Wilkinson & JT Olio
ABOUT THIS EPISODE
...now, injery CN CA work welcome to hashing it out, APOGASP Werwetalkd to the tech intovators behind blocked in intrestructure anddecentralized networks. We dive into the weeds Togot at Wyand how peoplebild this technology the problems they face along the way home, listen andlearn from the best in the business you can joined our racks episode. Thirty five hawshing it out. Iam door cory, very five, thirty thirty five is it not are was inthe last one, thirty, three pretty sure it was thirty. Four Aso episode three Xof how she got Ou COAACA, see which one ot is whenever we put it in Plos,preduction uh. As always, I'm doco, GA, petty and myTrustyco host is Colicuchet, say hello, everybody, hello, everybody, and today we are with J T and Shan fromstorage, a decentralized file storage company, and they are coming out with alot of really cool stuff lately in their news released in storage, and wewanted to get them on to talk generally about the difficulties of desentializedfise toage. As it's a something very interesting to call on.I've always found it fascinating and useful. As well as KINDOF, what their newfuatures are and and and how they got there, wherr things are going sowelcome the show guys you want to give yourselves a quick introduction as towho you are how you got in space and W at what storages yeah son, Hurs, Hey,I'm shon, Wilkinson, founder of storage. I got involved in these fun bit Cleenand Cpicurrencies in twenty twelve and fell in love and h throughout myjourney. I I you know, explore different parts of the technology andjust thought: Hey, desetralized, storage and kind of a big contedwatching make sens together and Kindof went down that that rabbit, hole andnever came out and at the end, was the storage, which is a decentralize and distributed Um. A COD storage blatform at allows anyoneto run out their actra hard drive space and on the other side, we allow peopleo store date on this network or focus on developers, but they can doa lot cheaper, a lot faster and a lot quicker Um. So an one we're privandandsecure. So that's US thatme and I'm jt H. I only joined storage h.just this year, so 'm grateful to be able to join sean onthis. Previously I was one of the early engineers at Mosey, which was an onlineback of company prior to Um. You know really people really believing thatcloud storage was even a viable option. Um We kind of helped push the narrativethere at mosy. I also worked and turned a google. I worked its space monkey,which was a kickstarted distributed storage platform with Um little home devices to compete withdrop box and h that that was sort of a a sixyear journey after an acquisition, andto, some other distributed sarageplatforms, and then I joined for a jdjust this year, so um I kind of have Um. I got early in Te, crypto currency,but Um. For the most part, my experience with distributed storage isactually nothing to do with cryptocurrency Um, more kind of on theacademic site Threreh's. My graduate research, so cool um sounds very usefulUm. So for this particular position. So Ijust I want to open up this interview with a question that I think is reallyimportant to just clarify right off the bat um most mydecetral's application developmenthas involved Um IPFs and I think it's important todifferentiate what you're doing versus what IPF s does, because it isdrastically different, but kind of in the same category of you know, h needs if that makes ense.So could you please just do that sure. So I think you know there'sthere's a lot of m people in the spas of who played around with IPFs Um, andyou know trying to highlighe the differences between them. You know theyboth allow you to store data but fror. My understanding, withibfs, there's, noguarantee that that data will be around...
...once Hou store it versus that's verydifferent on storage. You know you can store the data and have guarantees thatthat data is going to be there and available for you. So that's probablykind of the largest difference. I think if y ibfs is more kind of a do, you sentralize weight to kind offind and address lalls, but they're still kind of a missing pieceunderneath, which is you know, keeping the data around over time, which is youknow what we provide? So maybe JPJT can expand on that, but I think that'sprobably one of the largest differents h difterances between the two. I do. Ido think it's a good point that m IPFs and H storager. You know okay dicentralas storage, but pretty different Um yeah. I think that's a great point Um.I think. Ultimately, my position, you know just just off the bat is I'mreally hopeful and interested in. I P S' success file coin even made safe. Ithink, certainly with this space m kind of a rising tide raises all boats and you know, sort of the more the moremainstream essentialized storageis, the more uptaken adoption. It's going totake. You know in in in more UM professional settings, so I'm really excited about just sortof the whole space and you know doing what we can to help, but I do thinkthat storage is targeting kind of a different market segment than IPFs, andyou know the Filecoin ipfx s, IPFs extensions are mainly storage. Primaryobjective is to take the existing, centralized customers of cloudstorageand help move the needle a little bit more towards the central eyedstorage.So our primary focus right now is an s three compatible Um cloud storageplatform and sthre compatibility actually imposes a lot of restrictionsand a lot and makes a lot of design decisions for us, and so you know I pfswhile IPFs. You know, of course, if you've used IPFs you're familiar withkind of the imutable Hashes, that files are stored with it's something where you can pin fileson certain computers to help contribute to those files to the network. It'sit's a little more M of a community. I think in some sense and for us whatwe're trying to do is people pay US um destor their data, andso we need to turn around and make sure that that data is reliable. There's nopinning there's no um sort of you know like mechanism to allow people to choosewhich files get more redendancy. It's just a matter of. We want to store thedata with estre level, durability, estry level, performance of estry levelcompatibility and do it in a way where there isn't actually a data centerinvolved and so yeah WV. We've made a lot of differentdesign decisions right, like aicore files are mutable. It's it's a PALFbased system. Where you can change things, we don't have hashes thatidentify files that you can address and it's it's something where you know.Obviously, one of the one of the big issues right now with IPFSS. If you golook at the IPFS Hashes Sab readit, you know about half of them. Don't work,and I think part of that is is m. You know just based on it's. It's kind of a volunteer platformright people are contributing on IPFs their space and their uptime tovolunteer and Um. You know not not every IPFs note is getting paid rightand so filecoin, hopefully, will help address some of that. But that's justsort of like off the bat get go for us is storage. Note operators get paid,they get paid for get UPTIME, they get paid for a liability they get paid forstoring and returning data, and and that's at we need to do. We need toincentivize people to do the right thing so that we can provide thissthree level, type storage, so yeah kind of kind of a different M marketsegment. Ultimately, I think that if and when filecoin launches or madsafe gets their storage platform to a to a farther level, I think ultimately,that's going to be kind of a different storage product right. Like so there'sdifferences already with cloud storadge. There's s three there's placier there'syou know: Google has their nearline cold line, Google storage and there'sWasabi back blaze right. All of these have slightly different trade offsmultimately. I think that w kind of some of the other players in thediscentral ized rage market are ultimately going to be the forefront of kind of a new, a new type of storage that I'm reallyexcited about, and it's just kind of not what storage is focusing on rightnow. Torge is focusing on helping bring to centralization to the existing. Youknow exybites and Yatabites of data that are ultimately going to be storedin cloud data sinners. Unless someone...
...does something: okay, so olog before I,we go deeper 'cause we could easily and quickly go deeper into what you justsaid. I would like to maybe back up just a little bit and maybe providesome contexts as to why there's in need for decentralization in storage, wherethe problem exists currently and how things re, stored and and Davas inters,and how discentralization helps mitigate those problems and what thetradeoffs are associated with the two? Yes, so you know there's a lot of com.You know ways you can attack that Um. You know one way that people kind oattack, that is in terms of ideoloty right Um. You have these largecompanies Amzon. You know Licrosoft Gool, that TOR majority of the WorldTata- and you know we as users- want a lot more. You know privacy and securityand control over their our data that that's an interesting case for forpeople in users, but it doesn't really move the larger needle. At the end ofthe day, compaies people are spending roughly a hundred billion dollars,oncad storage every year, and that keeps on going up everywhere every year.No one saying I need less cloud storage, please and you know, there's a lot ofproblems associated with traditional problems right, so you, you obviouslywant. you know faster speeds to transeof that data you want to you know, spend you know as as less as possible.Um. You know again you you want. You know security, an privacy with that youassume that, but the problem is Ike. You're building out these, youknow six hundred million dollar data, centers Um, you know that's that's alot of capital that you have to invested into to building ose out andthe way that we we access and ose. The data is, is not in a sentralized manner,right everyone's, not in Rurale Nevada, where that data center might be Um. Youknow the Internet is a distributed, Ande centralized network, and so wereally kindo started from that segment to say: Hey, we store the data on aDISTRIBUTEDG, decentralized Internet. The architecture just looks more likehow people use an access that data and then as if by magic. You just startsolving some of the problems right. So if you want to bring costs down, El youdon't have to. You know, buy that fire suppression ystem, an that expensiveheating and cooling and the parking lot of the Datacenter, like all those costs,go away M and when you're renting out you know, people just excess our dryspace, but you know those costs exist in the Adison Um. You want it to bemore performant. Well, you know, if you're, storing the data- and ithappens to be three blocks down rather than through states. He way you know,then you can get a lot better, latency and perform. Instead of that and thentaking you know from a position of privacy and security since we'restoring ot Bhese on many untrusted evices all over the World Um, you knowthe we have to encrypt the data before. Even you know touches the Netlork, andso you know it's it's just taking a very fundamental different approach tothe problem and what you find is again that, since the way we access and usethe Internet is inte, decentralized and distribated way, you end up solving alot of these traditional Um. You know problems that people whave and so we'vereally focused on use, utilizing that that really extreme advantages and boilatingdown into a simple package in a simple service that people can use,and so we can start Kinda attacking that you know large hundred billiondollar market and is saying: Hey Ou'rusing. You know most likely, ms onUS three and you're paying this much. But if you maybe take five minutes toreplace a couple of lines, I can fig. You know we can save you. You know halfthis amount of money with better performance. You know, that's that's aobringer! U, and so that's where we're really focusing on is, is using theadvantages of technology and a simple in a package to make a a big impact onkind of the traditional clod market. So Yo, you mentioned s three early on Um,as kind of part of your solution, um you're not dependent on is three or W.where can people store these files um other than just like s? Three like? Areyou competing with historey like? What's going on there cun you clear,find that Yeah Soi'll make a comment on them and then jt, probably follo upwith o e some additional UUL information there. It really startedwith Um kind of V. Two of that I work so we've been around for a while we'vewatched, you know many iterations of the network. Um...
...version two over the network that wewatched Um last year, um in early two thousand and seventeen um scaled up toabout a hundred and fifty petabites o data. One petabite is one thousandterror bites and we had about a hundred and fifty thousand people um renting out there theyr hard rice spacean over a hundred and avy countries, so huge network learn a lot from it, butone of the things that we realize and learn from actually having this youknow live network in production. Um is that you know we. We had libraries thatpeople could use to to integrate with and they're kind of storage specializedlibrary. So it might take someone you know, hours, H or or or days to tofigure out how to antegrate Um. But if you look at the traditionalmarket, most applications are using. You know something like Amazon S, threediscor Dat, it's kind of the standor that everyone uses forgoodor for badright. It's this the standard that people use, and so we you know, lookingtowards The v three network and what we really wanted to change and make thingsbetter. We thought hey, you know, let's make this Amazon as three compatible,not dependent right. It's S. it's compatible with those those Apis. Thepeople who you know e are are are building applications like Canliterys,been a couple of minutes to get something working, and so you know wemade that change and it's it's really paid off. So far as we've had you know,partners h a d customers play around with ourearly version of of of the V three of the store's network and they're, justlike Gah took mew minutes, Um Tenegrad and that's that's what's reallyimportant, because if you look at the you know, decenflies and Distrubilandscape, there's a lot of people an a lot of technology. That's reallyimpactful, but you know it's it's hard to useright. Try to set up a you know, get someoneto set up a bit coin wallet or some ofthese things that certainly comealong way Um from from the early days. But it'sstill, you know very difficult and hard, and so what we really wanted. It do say:Hey. We want Ta Inpack, the traditionalmarket and we want to have a bing impact and the easiest way to do thatis not making it hard for people as much taking. The minutes is important, sothat's that's one design this reson that we made that's very, verydifferent from many of the other players in the space. So somethinganswer answer your question directly. We with ESTHRE WE MIMIC T E ys threeAPI. We don't use US three. We are a drop in replacement for sthree and what that is is a really good way to lower the barrier, injury forpeople who are already dependint upon that standard, which you talked aboutlike using things like esthree, so they can switch an entire portion of theirback in by changing just a few lines: Rof the actual code yeah, maybe evenjust configuration information only in nocode that'that'. That's a really goodM, in my opinion, a way to get people to use your network versus others andallow like people too into it, what's actuallygoing on or en even need to into it, what's actually going on, but wantpotential pervormance benefits or cost benefits of using your netbwerk versusas three I'm curious s before you said M, a lot of that. You said that I ldsay, for instance, I' Mea yacfas is the example here, because everyone'sfamiliar with it, the Hash that you receive, is based onthe content you put forward, and so you know that if the Hassh changes thecontent changes and you have guarantees around the data that you're actuallygetting. How do you D insense? You said thefiles are mutable within your system. How do you give guarantees like that ifyou can yeah so this is? This is actually goinginto kind of a much wider discussion about or overall architecture and kindof some of the decisions we've made. Ultimately, some of the decisions. Whatwe've made are potentially a little surprising and since the only thing we use Um, ablock Cain for is for background settlement and payments, and so, interms of you know, a lot of a lot of these cloud storage products. Thesedescentralized Clas torge products spend a lot of time talking about theirconsensus, prodn call um or they do things like these. You know they havelike a Mircle tree to construct the file system. They do content.addressable storage, like they hash the data, to make sure that you know whatdata you're getting because, honestly, you know these. These platforms arebuilt in ways where you can't trust anybody right. You just can't trust anycomputer that you're straying any of the date on, and so we've made. We'vemade a slightly different UM incremental decision again, our ourroadmap and our plan and our goal is to...
...kind of be a little more promethian andtake the fire down from you know: Olympia down to the masses of of thecentralization, and so we want to bring people steps closer instead of makingit a huge leap, and so the reason why I say it like that is because Um one ofthe things we've done is we've said you know we can probably get a significantimprovement in performance and a significance, reduction and complexity by making sometrade offs. One of them is okay. There there are computers that we want peopleto choose to trust within the system, and so by that what I mean is there'sthree different actors in our system: Theyre Storage, notes, Thei'rsatellites and there's uplinxs, and so we talk about this a lot and orrecently released ninety page white paper that took weight light reading, yeahjust way too many months of time on that one but yeah. So the the the stradge nodesare untrusted. Those are th. Those are the nodes that are the vast majority of actors in our system. Storage, nodeoperators provide their hard drive space and provide u storage data andthen there's satellites and satellites are run potentially by you. You can runyour own satellite as a customer of storage or you can use a satellite thatsomeone you trust has set up, and so this is the trade off that we made. Aswe said all right, we believe that it's possible for people to still get mostof the benefit of de centralized storage, Um and um are comfortable having anaccount on some specific server somewhere ore, specific set of serverceright. It doesn't have to be a satellite, isn't necessarily a singleserver for uptime. It could be a small cluster of servers, but the question isa trustboundry, and so a satellite is a small trust boundary that you arecomfortable, giving some medidate into, and so that's how we actually, weactually avoid. The White Paper talks quite a bit about m one of the one of the ways that we geta significant increase in performance. Um is avoiding coordination wherepossible. There's a a growing body of M distributed system's research, academic,distributed systems, research that talks about how coordination is one ofthe easiest things. Avoiding coordination is one of the easiestthings to do to get your r your system to be able to scale and continue toperform right, and so what I mean by coordination is m. If you have um H, for instance, bitcoin is a greatexample of coordination. Everything is coordinating and so um adding because everything has to agreeon this single global ledger. Adding more computers doesn't get you more.Through put you know, the amount of minors in in Bitcoin has increasedsignificantly, but the amount of transactions bit coinking processhasn't increased on the alternate on on the flipside S, three, the way thatAmazon has designed as three is very coordination afvoidance in thesense that it scales horizontally significantly other things that scalehorizontally are Cassandra or Um cockroach DB or spanner. Google, spanerdatabases, scales, horizontally and and horizontal scaling requires that youare able to I by adding more computers you get more through put and the waythat you do, that is you. You can't have every operation go through asingle global ledger and so by having everyone choose their own satellite tostore medidate on is kind of a way that we've partitioned the network intothese little truszones that allow us to avoid coordination. U, and so we talk about that, a lot in thewhite paper and then and then, of course, one of the main things that wewant to do is allow people to be able to use storage aciss it directly fromvery light weight clients like their mobile phones like there. You know webbrowser, their desktops and so in uplink, as the third actor in oursystem in an uplink coordinates directly with storage nodes and withsatellites for the Medadata that you have and so um it's it's a little bitof a surprising design. It's not maybe what you'd expect if you read aboutIPFs or or file coin, but it actually ends up being almost exactly theoverall design that a lot of other distributed storage systems that aren'tdecentralized to use like I I don't know. I think I think people havedifferent definitions of distributed in the centralize. So I probably shoulduse different words, but Um Luster is a very well known used in the academicsuabsolatly right lustre. It runs like fifty of the top one hundred fastestcomputers in the world. It's it's a...
...storage platform, that's distributedand Um. The overall architecture of of Lucstereis there's three components: There's clients, there's metadata servers andtheir storage notes, and so the design of storage Sr j Um is very muchinspired by systems like lustere. That makes a lot of Suyeah A and youknow put it into more. You know concrete Um example: you know, JT comeswith a lot of experience, you about e literature and you know hands onexperience with these. These many distributed h and just in generalstorage systems and we've come in. You know. Launching RV two network Um,which you know out class, is pretty much all the otherdecentralized stores, not wor, im bines in terms of scale and agnituded, and welearned a lot from that Um. But it's it's all about choosing. Youknow proper design that that works in Scales Um, I I won't name any names,but there is you know particular Storage Survey Story Platform that R lie O onconsensus, and you know users as experience issues. As you know, they'reuploading, you know a cat picture size file, a couple of megs and Um. You knowthey're getting L K six hours of uplow time right. That's know you might haveyou know the Nice Ideologicloo omponents in there, but if it takes sixhours, aphletic Apicture, you know, that's, that's not great Um, and so wereally wanted to come with something that you know works and scales off. TheBat Um allows you to get. You know performancethat of what you you you expect were better and then over time you know wecan make improvements incremental improvements on my design to make itmore trustless and more robust. So, forexample, F, if you really, if menability was really important to you,you could h, you know wride a rapper bfs in storch and you would have youknow your cakang eat it too, and so we really want to make sure there's asolid baselayer Um to to build upon that that works. Well. Um now O ot hereyou're focusing a lot on the user experience. So maybe you could stuck usthrough the the user experience of getting storage on the system in thethree peer classes that you've kind of outlined in your white paper. So youknow if I wanted to be an UPLIN node. I guess node might be the appropriateterm, I'm not certain Um, then what would be my path o that, if I wanted tobe a satellite, won't be my path that if I just want to be a storage client,what would be my path? It was he scent of models around each one of those peerclasses. That's a good question M for for an uplink, a an uplink actuallyis a peer class that were really describing to to match almost anyapplication that uses storage, so we're we're. Ultimately, we have an uplinkservice in our stor Adrepo right now, but we're working on releasing LibUplink, which is just a library that processes can y, can link against anduse Dan a lib up link we're going to rerelease our V two Lib storage, backedby Lib up linkond. So all of our language bindings will be just live upLincoln. So anything that uses that library where, for the purposes of theWhite Paper, calling it in uplink, so there's actually nothing to do tobecome an aplink node or an upling apere. It's just something thatnaturally happens if you're using storage Um, we do have an s threegateway that allows a a a computer to pretend to be the s threeOm Api. It serves the s three compatible inpoints, and so you can runin an uplinkn gateway that way m. But that's that's kind of the end of youknow, there's not really anything that anyone would expect that you do to bean a plink note. You wouldn't be an EPELINC note for someone else. If thatmakes sense, it's only if you're using the network Um to be a storage node. That's actuallyour next release. That's coming up a little bit later this year. Sorry, at the beginning of the nextYeri I should say um our storage note release is our nextbig release where people will be able to Um. Currently we have a weight list and allthe weight list is is we are currently using the certificate authority, thesign certificates, to join the network. We will strip that out eventually, butfor now we want to make sure that we grow the network at a kind of like a not a bursty rate. So we want ta meanher how quickly the network crows, but our next release is a storagenodrelease. You'll join the network. Yo'll you'll create an identity. It's a longlived identity that identiidentifies...
...you across. You know multiple upgradesand reboots, and then you will just configure it with m. You know kind ofwhere you want to get paid and that storage note will just start runningand start advertising itself to the network as a recipient of storageyou'll be able to configure how much ban with how much disspace you wantitto use and it'll kind of just do its thing, itd sort of set it and forget it,and it is really important that storage, node operators choose m computers orservers that will have good uptime. If a note doesn't have high quality,uptime or high quality availability, it will ultimately no longer get chosenfor new data and we'll stop getting income running a satellite is probablythe most intensive process, because the satellite does have a number ofresponsibilities, but running a satellite. Hopefully at some point willbe as simple as Um. You know installing our binary we usego for all of our programmings programs, so this is all statically linkedbinaries that are easy to install. But you know if a doctor container is, isyour jam we'll have a doctor container that'll make it easy too. He pointed it just a little bit of h, contiguration and a database, and youlet that run that will have an Admin, a Web admen consul interface that you canpoint around and and see how things are going, Um and so running. That will probably be abit more like running Um. You know some sort of service thatyou'll want it to be. You want it to have like a domain name that people canaccess it over and a few other things, but otherwise it should be as simple asthe inacentry running a service pointing at a day database. That's itso that's kind of like how you would be a satellite operator how you would be astorage note, operator B. Why would you be a sutlite operator? What is what isyour incentive? Aroun every operator gets paid, a storage note operator getspaid for the storage, but storage isn't the only thing that needs to happen inthe network, so the white paper talks a lot about repair and auditing, and soone of the things that we do N- and I know you mentioned you'd like to talkmore about a racia codes- There's actually a lot to talk about with Tracier codes. We've chosen eraciar codes instead ofreplication and there's actually a really deep argument about why that'sactually critical and vital quit roqes. You please to fine aracour codes forthose nd es that you' understand it. No great point in Araciar Code, so so yeah.Just briefly, when you store data, the question is: How do you deal with nodesdisappearing? So let's say I store data on the storage network and some storagenodes M, lose power or and asterate hits them or the storage noted operatordecides he hates us and weats, or you know, there's a bunch of differentpotential outcomes where you know a storage note operator mightjust decide Shehas had it with the software and just wants to install andleave so um. At any point we might lose data, and so the question is: What dowe do to make sure that all of the data that people have given us? We can stillreturn when they want it. Replication is a common choice. Youjust make more copies. It's Kindo, like the most obvious thing to do. Um Youjust in any incoming data you make. You know three copies: Five copies tincopies the V two network. We which m you know. As John pointed out, welearned a lot from thees vto network used a mixture, anitused replication,annoracior codes, but other systems use only replication D and Ur V. ThreeSystem only uses a racier cotes and what a neratior code is isinstead ofstoring data. As copies, we actually use a pretty interesting math trick,which Um I can explain in a second to break thefile up into multiple pieces where, let's say a file comes in, we break itinto forty pieces. We only need any twenty of those pieces to recover thefile any twenty, so it could be the last twenty, the first twenty, everyeven piece, every odd piece. It doesn't matter. Any twenty of those fortypieces will be able to recover the original file. So that's in a raciercode, there's a number of different Aracia code, Algorithms, but the mostcommon one that's used in you know making it. So you can scratch your CDand it still plays music. U Satellite Communication Bunch of stuff is anOugor them called Reed Solomon and so read Solomon kind of the way it worksjust sort of like at a high level. Is You know if you remember from mathclass in high school or Early College? If you, if you do Um, you know, any twopoints will identify a line right, and so, if you put more points on that line,it's still the same line. So any two points will uniquely determine a lineregardless of what points where those points are on that line an in the sameway in a quite a quadratic equation. Any three points will oumeekelydetermine a quadratic equation. Any...
...four points will uniquely determine acubic equation and so I' the same way. Any twenty points willuniquely determine a degree nineteen polynomial, and so, if we break takethe file break it into twenty points, twenty math points like where we justtreat them as numbers 'cause, it's just ones and Zeros. After all, we take thefile, we break it into twenty pieces. We treat those pieces as points on adegree. Nineteen polynomial, then we oversample the polynomial. We generateanother twenty points on that polynomial now any twenty of thoseforty points will allow us to regenerate the original file, and sowhat we do is we actually do this on a kilabite by kilabite level. So we canstream data through the network, and so this is how we do video streaming D andstreaming storage of log files. While we still get this property so n, wechoose some nodes, t storage nodes to upload to we take the file, we break itup into all these pieces using Rit Solomon, and then we store it on thosenodes in a way that now we can lose anytwenty of those nodes and still recover it. We don't actually use twenty forty.Currently, the numbers are actually something that is, you know we determine based on thedurability and the current characteristics of the network, butthat's kind of the idea behind aratiacoats an I guess. The mostimportant thing about a racia codes, I guess to point out, is the durabilityof the data is much higher using a raciar codes and the reason for that isbecause when we talk about replication, if you just make copies, let's say we have. You know you takethe data and you store it on a bunch of notes and Um. If you are only doing copies and youwant to be able to survive, you know like for n node failures. Um, you haveto have five copies right, and so now what that means is there's five timesas much hard drive space being used as the file that you care about andbecause there's five copies with Reed Solomon. With this twenty forty scheme, each piece is one twentieth of the file,so that actually means since there's forty twentieth, it's only two times todisspace. So that's called the expansion factor and the expansionfactor is significantly less with Eracia coats so m. What that means isit uses wayless banwith? It uses wayless dispace it dilutes. I it makesit so that we can afford to pay more to storage, note, operators perbite and it actually makes much highererability than replication would. In short, it basically makes t it givesthe same guarantees but more efficiently. Yes, that's ther's! So anySAPO, coing onears don on Tath side, ritoldand codes are Oldas as hell. YeTheyre from D nineteen sixties and there's been only lot a lot of new Arr, a racier code systems out there,one of the more interesting ones that just came out of Patten as a tornadocoat and one of the downsides to read. Salomon e thinkis that February right.Ah, the original fountain codes come out of patent in February and I'm notsure Tornado cocoe tornato codes should be out of patent. If I recall for myresearch from Luxu, I should look into that again. Yeah because, like reetSulleman, its encoding time is tremendously. It's like, I think it's on squared in order to do encoding or notes anloging uncoding andto decode its on squared, whereas Tornado codes are pretty much just and log natural og en it's it's pretty.It's pretty quick so like what would take like a sixty, meglife file wouldprobably take. I don't know thirty thousand seconds to Um to encodeand maybe like thirteen thousand a decode in retalement. It's it's! It'slike four seconds to to one second in in tornado and so like H, th, the tradeoff theres. Of course, the the length of the raciacode is higer in Tornado,but, like you said, it's still way less than replication and way less thandoubling the size of the file. So I'm kinda interested in what made you gowith the read Sullimon route, rather than some of the more modern Oraciocodes that are out there. So I think that's a great question M. for the mostpart, I've kind of just avoided fountain codes. Just due to patentencomberament and kind of my personal philosophy, s did not go read. Patents'cause, I don't want. I want, would rather preserve plausible temyability, but in terms of Resalmon it is old, umh it could be faster, but it's not like it's slow, read, Solomon, a good and Co.A good decoder can becode at three hundred neabits a second or make abites a second which is usually faster than the linksthat you're talking Bou. So Um. You know we. We think that overall, themajor the major performance requirements for us are actually interms of Bamouth and...
...throughput and latency, and so the ReedSolomon thing just isn't below ust hanging through, like you know,replacing it with a better eraser code, St Scheme Um. This one works. It's we're we're convenientlyconfigured in such a way that it's easy to replace, read Solomon with somethingbetter as soon as it becomes our main bottleneck. But for now I mean T, likeI said three hundred megabites. A second is certainly not as fast as itcould be, but it's not the main bottle neckfor competing with esthre a that okay. So what Um? Just so theaudience is clear. What is the Um Maximul loss that you can suffer on afile and be able to recover it using the Racier cods? It depends on yourconfiguration with your Racer Code Right now. I think our code baseddefaults to m twenty nine, eighty, five or maybe twenty nine ninety, orsomething like that. It's not quite the same thing as a two x expansion factor,it's a little more than two x, obviously, but and what that means is.If you have eighty five pieces, you only need twenty nine of them. So so,what's that Egh, five minus twenty nine is and e fool and the full a racier cothm and the fullerraciacote. So you can't just so even like you need twentynine pieces, you also need the full oracier code in order to recover thosetwenty nine pieces correct. What do you mean by a fuller, racier code? In otherwords, when you have the file itself and the Eracer Code is basically taggedon to the enetholysis considered, the eraser coded pieces are actuallyall you need. So when I talk about these pieces, you only need twenty nineof the eighty five pieces. Those twenty nine pieces are the output of theOracia Code Function. So that is all you need. You only need those twentynine pieces to recover. The original file is Um. Is this configurable for the end user,based on like their R prevraces on how much they need their day? They'reavailable yeah, so our intention is to make it so that people can choose theirdurability and then we'll have an estimator for given durability. Whatthe Best Reed Solomon Choices are now the trade off for that is going to beprice because the more storageredyour data cross it's going to cost more, butyea we do want to have a sliding scale. Where you can say look you know, Idon't really care about durability so much. This is something that I'm onlystoring for ten days. So I don't want to pay a bunch for durability or youcan say, IAM planning on saving this, for you know you know the next fiftyyears and it's you know my kids baby photos, so I want really hyderability.So, yes, there's a number of different choices you can make on that Pectrum S. let's talk about privacy, then Um so yo. Basically, you you pitch theprivacy as one of the key things usin a certificate authority to get peopleonline. How are you handling encryption and whohandles that? I think it's is it the uplank Lod, that's doing allthe encrypting and Y and how? How do you h? What what is your scheme forhandling and delegating the access to a FI, sur, great, so M, so right? So theuplink is what does all of the encryption and we have a reallystringent policy, which is that none of the other systems have any access toany unencrypted data. So the satellite doesn't have access to an unencrypteddata. The satellite only stores encryptdid Meta Dava and it doesn'thave the keys. If you lose your keys with storage, you lose your data and sothat's kind of a weird user experience straight off or people who are used topassword recovery screens but um. We think on balance, at's, probably thebetter choice from a privacy perspective. The encryption scheme thatwe use is configurable and we default to a SGCM, which is one of the M Nof,the new authenicated encryption schemes. I guess it's not terribly new, but Um.But honestly, my preference is actually one of the H, Daniel J Bernstein encryption methods, which is the secretbox. Encryption method, which is m polly thirteen, oh five and Chacha um encryption and m anyway. What you cando is you can choose your encryption key? You can figure your uplank withyour encryption settings and it encrypts all of the data before it evergets. Read Solomon encoded before it ever gets, int anywhere, and so theonly way you can retrieve the data is, if you have that key with the UPPLINK.That's doing the retrieval, because this is important. You know we wan tobe able to support a lot of the functionality that, as three does,which is like you know, being able to share a file with someone you mightwant to be able to share. You know a certain different delegated accesspatterns. Right NESTRY has a bunch of rich um permissions,...
...so um the UH. The goalf for us is to actually make itthat we have um encryption, that's hierarchicalencryption, so our encryption system is based on. I think thirty, nine hierarchicalencryption I mean bitcoin. Did you know hierarchically encrypted wallets sothat you can have a sub wallet that you share with someone? But you only have amaster key, we're doing the same kind of thing, butbased on encrypted path names, so every path element so like by a path element.I mean, if you have like you, know: music, slash, Um. You know ArianaGRANDA, slash! Thank you. NEXT OT MP, three right Um the path elements are music and then aron a Granda, and then thank you next that MP, three, those are all separatepath, elements and they're separated by slashes, and so the goal is. Maybe Iwant to share with you just my music folder, but I don't want to share withyou the full bucket, so the the encrypted path for the music folder. Iwould give you that prefix and I would give you an encryption key that allowsyou to access Um hierarchically encrypted, a keyarivation scheme for just everything under that music colder. So there's athere's, an issue with Gi. Thirty, nine in terms of how you can you can use aspecific subset of of? U Like Sub Paths, to the regenerateparent paths. Is that an issue that can come with, like maybe potentiallyexposing access to files? You don't want peopleto have well, so I mean I guess I guess what I'd say with that is we're notactually using. Oh, you just want to give like an analogy. ofout things workwere were our scheme is inspired by. He erthere had to be some differences, andso, ultimately, in terms of like okay, just in general, security,vulnerabilities and stuff, like that, we think that we've kind of got itcovered, but that's really hard to say for certain Ned. Lo, like the wild exposure and people trying tobreak things before you have stronger guarantees around the security model.How you do things we need a lot of eyeballs and we're also getting asecurity audit, so we're getting least authority who you know that's Zuco'scompany prior to Zcash Um, to do security. Audit Forst after we're alittle bit farther along speaking about it. Your system does audit files correct,so it makes sure that you have some sort of uptime garetee and has some ability to guarante that the filbeingserved is actually the file. That's that's that's supposed to be served. Isthat correct? Yes, Li Yo Ha. How do you do that? What do you? What is, what isyour schemof for verifying that a file is the right files if you're not hashaddressing it, although you might be, I just don't know the details on that yetand how do you? How do you know that the fowser actually being served toanybody who requests it and not just the person who's doing the Validat? As a great question, I think I think some of the answersthere are well. So let me let me just start from the beginning, so thathierarchical encryption scheme does we are using authenticated encryption,which does both encryption and and signing right. So you know, if you can't, you can't Um. The data is,is written in such a way that, even thoughit's not a hash of the contents, the file isn't named after a hash of thecontents. We have a hash of the contents, so we know that the data iscorrect, because the Hashes are all the way included through the encryptionscheme. So that's how you know that the data thatyou're getting is rent um in terms of Um h. You know, auditing in repair. Ourauditing and repair system is Um. W We've made a number of differenttrade offs both on those fronts. Auditing is, is written in such a waywhere our goal for auditing this is this is actually, I thought, kind of aninteresting sort of trade off that we made most of the auditing systems that arewritten about for discentral, ized storige and I remember Atalic wrote ablog post about you, know kind of proof of proof of retrieval. I think where,basically, you store a Mircle tree of pregeneratedchallenges mean there's actually there's actually a number of differentUm papers about a proof of retrievability that include ways ofgenerating um challenges ahead of time, and you havea finighte number of challenges, there's also some homamorphictechniques that allow you to have an unlimited amount of challenges and whatwe're doing actually is you know this? We decided that our auditing system itsprimary goal- I is actually like this is kind of a big shortcut M forperformance reasons- were actually Um, not using our auditing system to findif files are bad d. So that's actually...
...a really interesting difference. Mostof these proof of retriete ability schemes are trying to figure out if thefile is retrievable and it does it by random sampling. So it's assuming that you're not going to audit theentire file every time what you're going to do is you're going to auditlittle small ranges of the file and probablistically. You can be prettyconfident that the file is there completely intact, because the storecan't predict what your audit would have been until ha store gets it, andso this sort of sampling kind of incentivizes the store to keep theentire file, because it can't predict what the oddit's going to be about, andso it's the sampling process that ends up allowing us to be pretty confidentin a general proof of replication scheme, that the file is there withoutdoing a lot of work, and so we're actually going one stepfurther and we're saying the question is: Is the storage note good? Is thestorage note playing by the rules, and so we have a list of all of the filesthat we believe a storage note has on the satellite. The satellite knows whatfiles a storage node should have based on the satellites, Medadata Dso, thesatellite is going to consider all those files and then consider all theranges within those files and then do random sample audits on randomlyselected files, and so the goal of auditing is not to determine if filesare bad. The goal of auditing is determined if a note is bad and if thenote is no longer playing by the rules and keeping the datat supposed to, thenthe node is as penalized and ultimately ejected sooauditing actually doesn't care. Somuch about the data correctness, the thing that auditing cares about iscatching noses with basically just sampling, Um spot checking. Theauditing system is just spot checking nodes to make sure that it's good thething that actually checks to make sure that the data is correct is our repairsystem, so our repair system, if it determines that a note, is bad. Where anote has been offline too much, which is an interesting separation. We thinkthat most of the data loss that's going to happen in our system, is going to bedue to notes going off line and not doit, not due to nodes being a corrupted or mutating data thatimagine shurn is much more of a problem than like Tanar nincrred recruption.Exactly right turn is going to be our biggest problem by far, and so we havea very high incentive, a very strongmotivation to care a lot about churn more than anything else, and so ourauditing and repair system is kind of all predicated around. We need to beable to quickly determine which nodes are online and how long they've beenonline and if they're likely to come back and if what happens is if it turnsout that you know ten of the forty nodes that store a specific file areoffline and, ah or or have been marked bad by the spot check auditing system.Then we need to repair that file and repairing the file is going to involveactually a mirkle tree of Hashes, and so that's where we're actually able toconfirm the file is correct and we have the pieces that we need. We can do theReed Solomon without doing any decryption and Um recover the original pieces and replacethe missing pieces on Ta new storage notes. So that's something that thesatellite does to make sure the dateis good n. The last question you asked was:How do we confirm that a storage node is um playing by the rules for uplinks aswell as satellites, and I think that's that's an area where we still have alittle bit more research and a little bit more to explore, but the plan sofar, if that becomes a problem, is to just consider statistically reportsfrom uplinks about storage nodes, not playing by the rules. The goal is: Ifyou only need twenty nine out of eighty five pieces, most of the nodes aregoing to be. It's going to be very easy to recover your file, and so the numberof nodes that are going to be Um, you know malicious, will be quicklydetected by a bunch of upplings complaining. I think, what'sinteresting about the way that we've, you know done this Reed Solmon thing is,it turns out. We don't actually depend on any specific pece, and so what thatmeans is that our earth are speed. Now only depends on the fastest respondersof any request. So we actually, when we do it, download we overrequest piecesmore than we need and then we're able to return the data to t e the user assoon as the fastest responses have come back, and so, even though our storagenotes have highly variable performance and we actually get the best of thestorage note, so our high variability were able to turn into a big strengthbased on that architecturae decision. So, what's H, let's talk about some.The common attack fectors on kind of decentralized systems. Here um like Wwh, what about like Um? What, if somebody throws up a ton ofnodes and gets good reputation on them and then starts manipulating thenetwork like how? How do you mitigate...
...some of these like or Sybil attacks orsomething eclips like these are all common things? What kind of threatmodels are you considering in what what what ones are still kind of in areas ofresearch? And what ones do you think are not an issue yeah? So that's agreat question: Um we off the bat decided that our nodes would actuallydo um at very least proof of work to join the network. It's kind of it'skind of sort of a dumb side thing that we're just making nodes.Do I it it SIMP. It removes a bunch of simple attacks, but it doesn't reallydo anything about a a motivated attacker. So if someone, if someone hasa bunch of you, know time months worth of time, can generate these IDs forjoining the network, they join the network. They just KINDOF, hang out onthe network for a long time and then start Um and then start trying to manipulate thenetwork Gat for our original design. That was actually kind of a problem. Imean you're attacking amazons business model, thats three kind of a little Bothere so like they have the power they have the ability state actors do aswell if they don't like what you're doing for one reason or notherthat'srgh. So so the question really is: How do we make that? It's not reallysomething that we can. I mean again. This is the same thing with bitcoin.Bitcoin has the fifty one percent attack, which is you know justsomething that hopefully is is too expensive for anyone to do is kind ofthe best solution that you can. Even then it's identifiable like the wholenetwork because of its public andsensus model Um it. You know, th, there's been a fiftyone percent attack on Bick Uin and it was successful and it lasted for alittle while and Um. You know it didn't actually have any negative consequences,nobody as an to capitalize on it, but in this particular system it canactually attack your straight up business model here so yeah. So so thegoal I mean yeah. The question is how how how much work would someone have todo to be able to attack our network and ultimately, a state actor is going tobe able if they are motivated. therethere are a lot of attacks thatare going to be very difficult to deal with right. Ok, if m we're storingeighty five pieces of anyone's file- and you know we looking at ourstatistics- storage nodes have a certain. You know. Failure rate there'sa certain term rate, but all of a sudden, the entire Internet is shutdown. Yeah the Datas not coming back until the the network's Backe Upright.So there's a number of different things that certainly motivated powerfulactors can do. But so the question is in terms of our threat model. We wantto make sure that we are a roblessed enough against the things that are thatthat are within reach, I think of of most most attacks, and so one of thethings is we were tha. We were kind of vulnerable to this problem of ofsomeone spinning up a bunch of nodes of in our early draft period of the WhitePaper. We we sent our white paper off to a bunch of reviewers and and one ofour reviewers graciously was Um, the actually the author of Cademlia Um, Terma Mamonca, and so he did afantastic job, reviewing our white paper and came up with animprovement actually, which is part of our system with Cadenlia. The goal isto make it so that the storage nodes have to provide. Imean it's: It's kind of easy to spin up some storage notes that just sit on thenetwork and then start mucking with things it's kind of hard to providepedabites of hard drive space and then start mooking with things right, likethat's, actually a really expensive thing to do. If you're that motivatedto provide our network with petabites of good hard drive space, then yeah,that's Kindof, really hard to defend against. So that's just sort of a problem ingeneral with this disentralized torge platforms, and we believe that that'sthat's unlikely enough of a problem to not worry about that too much someonebeing able to spend headobites. You know supply petabites of data to thenetwork, only to just mess around with people and so the goal is. We actuallymade a change in our architecture based on the feedback that we got, which wasthat condemn e there's, actually two tears with Cademlia. The DHT THAdistribute a ashtable we use and you can't participate in Cademlia untilyou've passed enough haudits until you've proven that you have hard drivedspace on that node, and so that's a way of making sure that these aren't. Thisisn't just a civil attack. If that type of thing happens, where immune dicibileattacks are imune t these type types of things, because it's actually reallyexpensive to get a node that has enough reputation in the network to be able toUm nend up, manipulating or denyingservice to someone with a data, and that's it's also important to get. Youknow the scale Um and get as large as possible, because,as e network grows larger than it becomes much more difficult to pull off theseattacks. Definitely, and- and so I guess theQuestionan is, there are certain...
...major system, so you say like we're:replacing data centers, six hundred computers whatever, but like most ofthose systems that require that kind of storage pace M. also raquires a certainamount of guarantee and a certain amount of security against attacks likelike we just said to an extremely high five nine degree. Do you feel, likestorage will be a direcorplacement for them, or is this more tuned towards acasual small business audience smoll temid? I think I think that Um this is this is this is an observationjust sort of from my experience at Mosy the hardest part at mosy initially waswh n when we were doing so. Mosey was an online backup company and it was one of the first onlinebackup companies. That said we're going to stor your your mostimportant data in the cloud, and so it was actually really challenging toconvince companies that what they wanted to do was take th ir. You knowtheir most important data and store it off premise. Um. You know this was youknow back in the time when, like iron, mountain nd and some of these, likehuge backup companies, you know, would bring tapes and then they'd like Trupyour tapes off to to make your backups just because people really cared aboutthe backups and they weren't really sure if they should trust someone juststore their. You know your backups in someone else's service. Is that that'ssafe to do is that good and kind of Mosey's whole premise was yes, it'sfine encryption is good ut, we constore the data. You got to trust somebodylike you're, trusting your tape, provider and so M. Ultimately, it was kind of afight Um and it's kind of laughable. Now thinking about how every businesshas moved to the cloud everything's run in Gooe clod platform or Azure or NS,it's been such a dramatic change in the last ten years in terms of what peopleare willing to accept and willing to adopt, and so ultimately for storage.You know this: The Shift Tod discentralization m off the bat it'sgoing to not be something that will be able to convince everyone. Is a goodidea m we're definitely going to be talkingwith earlier doctors at the beginning, but the goal is to make something thatis high quality enough and m performing enough and reliable enough and secureenough that people wil, you know it'll turn heads and people will will startto really open up to the idea of dissentialized storage and not saythings that you know the sorts of times that I say all that you know. I hearall the time like. Oh wait, so you're going to just store my data under myneighbor's bed. Well, it's encrypted right, and so it'sit's. The same kind of conversation that that you know mosy had ten yearsago is this is a new paradime. The paradime has enough benefits that weshould consider it um and UM. It really just is going to takesome people jumping in and again it just goes right back to what Shansaid,which is the larger the network, the better this is going to be I'd, becarious about so part of part of your early adoptionis going to be, as always, a new brand new technology that distributes thingsis going to be sometype of ILICID media. How do you how you approach that subject? Are youeverything's encrypted, of course, but are you breaking up the files of thearratees? You store all of the erasia codes and a single place as as Tosugaracross people. So not a file isn't completely stored on a single mode sthat people who are using your services, just like hosting stores Media Aryo,basically can't ever be complicint in any type of storage of ilicet media.This is a great question. Yeah. We got this question all the time at Mosyactuaally about whether or not we were storing. You know what we were storing.We got this question at space, Monkeyala yeah, it's it's! Basically, it comesdown to you know saying we use. You know the bestencryption possible to protect you from privacy. You Kno Tod, protect yourprivacy with your tax documents and your baby photos and and- and you know,t those the sorts of things that are important to you. It cuts both ways ussaying that we can't access your data means we can't access your data nd, andthat means that we don't know what people are storing Um in terms of interms of you know where that file is stored because of the Reed SolomonNothre, there isn't a place where the file is stored in its entirety,anywhere. It's broken up with this mathematical trick after beingencrypted and then stored in little tiny pieces across all these nodes. Ithink it was either Shan or Ben or R CEO who said that were Engochan. Iguess I don't remember. Someone said what we're doing is we're makingencrypted sand and spreading it across an encripin. What how he? What even is a file likeit's just it's all math and from this...
...math you're able to get back your datayeah. I think John John said that one Juhnquin he loves that Analogi Yeah, I mean it'sit's you know, there's there's many different aspects that, but we really want to be a platform that that gives users you know, control over their dataUm and where you know th the Datas incrypted, the users have the keys. Youknow where Um you know we have a braacher get hacked like it doesn'tmatter 'cause we don't have access to an ither ato, which is a very differentsense from you know many other Um. You Know Cloud Sword forviders, where Umsomeone could get access to that data, whether it's just a malicious entity oryou know a government um I mean we. We know the the stories of you know howthe the FBI wanted. You know backdoors and and data from apple and othercompanies, and you know we- we take a very different approach, right iis,that we want to have a a secure platform head ED space. We, you knowopen source, all our code, so you can verify that there's, no nothing fishygoing in there and we just kindof want to be the kindof the neutral. You know, Sweden, you know data providers that you know doesyou know best up to protect you in your data um and in like he said you knowsometimes that that cuts both swaves. But you know we really want to be onthe side of of our users and our customers. First Um yeah, that's great! So speaking ofusers- and I know we're running a little long here, but I have actuallyseen more questions. I'd like to bring up Um. We can be brief about this. How do yousee people integrating your system with decentralized application, specificallyatherium smart contracts, O that'? That's a good question. So Um,like we said before, the're Um, you know our focus right now is is morein terms of the h traditional cloud platform. You knowthat already exists and people use its scale O, but we do have h many other people that are, you knowin the build stages of their technology. That are looking to use US as as aKindo distributed DA. A platform like Docai is one on m is another, and sothese are, you know, still in the very early stages of their developmentprocess. But a lot of people are looking to use us as kind of that thatDey delayfer applications Um I've seen you know a couple of examples: o people who have bill disturbateapplications, UM, adding an air quotes here, and you know the the applicationwill be running and then suddenly, which you know happens every couple ofyears. You know Amazon S, three will go down and take a quarter of the Internetwith it and that application as well and everyone's Jus like Oh. Why didthis decentrualize and distributed application? You know go down and thenyou realize, oh, they were destoring. All T e, the data on a centralized Um.You Know Cloud Score srovider. So we're really. You know youesil platformfor kind of these distributed H, andecentralized applications, but it'sstill stillwere very early in terms ofdevelopment. So a lot of people are integrating and and learning the toolsets and building their platforms, but I think it's it's a bit it's very earlyon in the stage of develtment yeah. So specifically, one of the useful thingsabout IPTESSIS is content ad rest and I can fit that content address within atwo hundred and fifty six bit register in Evm M. is there an analogous thingthat I could do with that on on storce yeah? So, hypothetically, you know, interms of that example, you could um since there's a lot of intigrationswith IPFs. Already you could Jus, essentially use storage as a back end Hfor IBS o. Then you would be able to Reu some of those same intigrations andyou'd have some guaranchees that the thfils would be available in there andthereud just be stored on storts. So you can use kind of existing layers Umto essentially Um. You know, Kindo have your Takin eted too Y 'causeif. This isbecoming kindof like an open standard almost, but the facto I guess you couldcall it IPNS specifically so olting that wrapper, like you said earlier,would be extremely valuable to anybody, because really, what you are is a backend for the storage mechanism and the addressing system should be independentof that to some degree yea. Well, we...
...have if people are interested incontributing to our platform. Some of our open tickets are descriptions ofthe IPFs Gayway that we have planned a is not on our immediate Rudmat for ournext release, but I'm an IPFS gaeway in a number of different directions,actually is something that we are very interested in doing yeah I mean that'sthe cool thing about Um you w distributed systems and open sourceright. You you have all these. You know cool projects on the space like IPFs oror Valquin, and you know we're really trying to make a dent Um. You know bit for ideological reasons, and youknow the the large cloud providers- Ammazon Kno, Google, microslott andwe'rewe're kind of all ialistically alline to to you know, store people stayed in a in abetter way, more secure way and more private way more performant way. Butyou know all this cut is up and sort and o wreaking to write. You know,integrations and and tools together, so to really uh have a a unified front.incode against these. You know large providers who you know exist thanpeople use, but now maybe not have all the the interests in heart and the focus on the users that theseplatforms- you know the platforms that we and others in in the space aretrying to build. He E, I really think of the iser ser. So so I guess the lastthing: We've kind of skirred o past this issue and you've mentioned it afew times and I feel like it's something that t does deserve at leasta few sentences on you say people get paid. What does thatmean? So if I'm using your service, I'mgetting paid for putting up a node? What is your, what is your model forthat? Just just to clrify it Ouf of the audience. You want me to take this oon or youwant to take a j, no go ahead. That's that's! Why you've been recking on yeah, so the the basic you know premisethat that makes us work. Is You know people have extra. You know hard driespace on their computer, that they're not using and they alreadyhave a stable Internet connection, and so they co. You know download thisprogram. You know, take a a few minutes to set it up and us they'd go and letit let it run in the background, Um and earn money if fere storing that dateover time and serving that data frend requested Um, so users are essentiallygetting paid. Um, our native token storage, stor, J Um for essentiallyproviding you know, hard drive space and banned with Um. So it's it's kindof variable what they would earn based on h. You know the the stability anduptime of their nod, how much ou know dayto their storing and how much banwith that they're providing Um. But you know we're we're in a providing incentive and a directincetive for for people to to runt out ther hard, wride space and actually Khave some benefit for them. You know participating in this network whichincentivives make things, make things work Um, and so that's really importanta departure for some. You know other Um, you know cloud store platforms, thesurvey clous sports, paforms n in the past, where people have said okay, you know, if youshare out your space, will give you more space on that outwor on on thesnotwork and that's just Kinda. You have apples, and you know someone trades, you moreapples. You know you, don't you don't come out any better, but if you can,you know, take those apples- and you know trade them for o k ow something ofof some tangible value. Um. Then you know it works out a lot better,so um highly varia in terms of Um Realy Cou make based o kind of your setup in your note but m. At the end of the day, it's it's going to be a marketum for for people to participate in. I think that's a Wi Rad rap. Thisepisode up as always I' like to ask her. Our guess is: Is there something thatwe should have asked you or you hoped we asked you that we didn't get aroundto no, I mean personally, I thought thiswas a great conversation. I wonderful questions very, very unsightful. I Iappreciated the conversation yeah at's Great. I I would like to highlight.Well one thing: I really recommend y your listeners to get a storaged Otiosto R J that I ow and Tik. Look at the White Paper become a storcenodeoperator and play around with their tools andlibrary, or maybe make tsome contributions towards that IPFs skatewhat we talked about so plenty of stuff where people lookaround and do if you didn't get enough during this conversation, O standawltos is, and Eur shownotes and Happ have...
...your lawn as always as Um. If you likethis episode audience, please link it share it onto it or share it onwhatever media platform you you enjoy yourself contact us, join the slack and tell uswhat w t you liked o. You didn't, like anything, just share it as much as possible, so peoplecan understand a little bit more about how distributed storage ordecentralized storage works with white useful and how storagees is solving tat.AU, Thaxgsp, yeah and y guys will be scoring your date on onthe a outel. So yeah and t els has always fall of uson twitter at hashing. It out. Hod and ory is at Copetti, and I am at Koncuchethat COLLIMCUSC IE UPSID.
In-Stream Audio Search
NEWSearch across all episodes within this podcast