DAPPCON 2024: Decentralized Storage Landscape in 2024 - The Rise Of Native VMs - Garrett McDonald

Transcript

hi everybody what's going on um beautiful day glad to be here uh my name is Garrett I um I run a fun called permanent Ventures and we focused mostly on like decentralized applications which are set up to exist forever and have uh yeah just be very long lasting with the right token economics and incentives such that you know ideally the projects will Sur survive the founding team um I just think that's one of the most interesting um you know aspects of crypto is like you know the ability to create things which survive um yeah beyond the creators um which isn't as often the case in traditional companies so yeah um today we'll be talking about um yeah decentralized storage landscape um

specifically about VMS that are emerging um yeah so start off just giving a little bit of a you know overview of all of the competitors throughout the past that have you know um you know started decentralized St projects some of them have pivoted into you know decentralized compute and others have become a little bit too centralized to be considered anymore um but uh you know it's a it's a space that I've personally been following since like storj and um you know swarm and some of the earliest uh earliest attempts um but uh yeah today we're left with the kind of most popular four um r filecoin e storage swarm and of those the two that are um you know most uh most used and um that are actively

developing their own virtual machines um are just are we and filecoin um so those are the two that will'll compare mostly and um yeah so the respective virtual machines that they're working on um R we uh launched a test net of AO um a couple months back um which is a active oriented message passing Network um and filecoin uh also launched a you know actor oriented message passing Network like a year ago in March and um yeah it's really interesting to see the I think in particular like the convergence of the two networks uh in terms of their storage offerings in particular because of what a more generalized virtual

machine enables so we'll talk a little bit about that too um there's a bit of background to this talk which you might be interested in U if you're interested in these two networks in particular uh this was like a similar kind of talk just diving into um the protocol differences cuz now we'll just talk about the virtual machines um and there's actually a lot more that's different on the protocol side um so yeah can recommend that as like a you know attempt at relatively you know unbiased comparison between the two and there'll also be a link to these slides at the end so um but yeah just a little bit of an overview on you know what was discussed in that talk was um yeah um you know these are like if you really

boil it down like the key differences um so there's everything from you know obvious like language implementations um to um you know the validation proofs and um but the the main differences that are important to note is that you know filecoin is at for its intentions are to be like a you know like a a decentralized uh broker of storage so then you have you know sellers and buyers of storage facilities and you enter into time based agreements on you how long you want this storage for and it does great at that R weave is a protocol which you know has an endowment built in which guarantees that you know if you upload something to the network it will pay out rewards to miners indefinitely like in perpetuity ideally

making some game theoretic assumptions and so on um that the storage will always be there um yeah so there's a couple nuances there but yeah uh you can look at that to to dive into those um so yeah um the main the main topics interest for today's talk yeah virtual machines and the applications we can build on those um so yeah main thing important to note is that you know um uh yeah I think the convergence of storage offerings is pretty interesting like so R weave as a permanent only you know data storage level layer through the use of a virtual machine can offer uh term-based storage for the first time and filecoin on the flip side through the use of a virtual machine can offer uh Perpetual storage so you know the

offerings converge and that's pretty interesting um and the because of of each of those initial protocol level tradeoffs you know how virtual machines are implemented is uh is obviously going to be going to be different um but yeah I think probably the most interesting uh use case for these networks is going to be like um yeah autonomous agents and AI inference um in a truly decentralized way because of the nature of um each of them to run like a you know arbitrary levels of compute so um yeah the uh there's a big uh difference here in the two kind of virtual machines that are initially supported on each of the networks uh filecoin virtual machine is focusing on making a version of the evm

um that runs natively so then you know can easily de Port over G5 Primitives and so on and so far it's mostly used um you know for like fill like file coin reaking and lending to miners so then they have more liquidity and access to um you know being able to scale up their mining operations without needing to put up as much Capital up front which is great um web assembly though is in is like natively and inherently like just much more flexible in what you can do um yeah especially after we get over this uh initial limitation with web assembly one um only uh allowing up to 4 gabt of memory so for instance if you wanted to run some kind of language model inside of uh web assembly under AO or under fbm uh once they support that um then you

can uh yeah now right now you're limited to like a llm which is you know slight less than 4 GB right so um it's not as interesting so for example like if you guys use chat gbt or something like that you'll want to be using a model that uses much much more RAM than that you know maybe up to like 30 40 50 gigs so um yeah once we can run those kind of more serious models inside of um this uh like you know decentralized compute Paradigm of uh these protocols leveraging the data stored on each of them it's uh you know really fascinating what we can do um so something that R we are working on implementing um with AO is like uh yeah that uh that 64-bit memory space which would be really cool

um so yeah both of them uh AO and fvm are you know VM agnostic uh fvm are starting by you know implementing evm um but an AO are starting by uh making wasin the default uh but the idea with both of them is to make it so you can plug in your own VM um whatever you want to use and that will be make it very flexible in terms of what can be run on the networks um both networks um yeah use an actor model so you know basing on message passing which is uh you know great for like massively um parallelized compute in a decentralized way um yeah and so I thought I'd put up like what is

the actor model here but basically it means that you know each uh you know node on the network or uh agent that's executing code code is communicating with others just by message passing and acting based on when they receive a message and send someone back so that way you can run like massively parallel Computing Paradigm instead of having like a monolithic State like uh we're used to with most with most blockchains today um so where they start to get a bit different is um yeah uh is kind of like the intentions of what's being built out first so yeah in fbm you have the FM um whereas in AO you have this uh interesting like SSI model where you have a kind of initial image that you

boot up from and then uh there's different roles within that that uh can be assigned to the node that you spin up um so you can schedule tasks you can do compute um for instance some compute units will be good at CPU some will have gpus attached um and others that are meant for sending messages um so yeah um and then with yeah AO of course uh it's you can yeah you can think of it as L2 for that settles on R weave um which you know of course inherits its security um and yeah so I think I just said that but um yeah evm first versus message passing first but eventually you know the convergence between these two

networks will probably be even stronger uh and more prevalent when you know um you know AO will also Implement like a an evm and fvm will also Implement like ways of doing message passing so um right now they're quite different uh but you know ideally if they both uh materialize in like the visions that they have then it should look pretty similar in the end which is which is kind of cool um I think it's very validating for each of the projects that you know they want to go in such similar like end directions but um yeah um both of them will be able to you know manage storage deals through virtual machines which is really powerful for yeah implementing the you know Perpetual storage on um on

on file file coin and you know temporary storage on AO um yeah doing compute on data that's natively stored on uh each of these uh each of these blockchains and uh storage platforms is really interesting so you know having native access to you know large amounts of data um and being able to do compute on those uh without needing to worry about transporting that over internet over long distance and having the nodes that do that compute um maybe even in the same computer the same data center um can be really really good for um yeah um like U you know optimizing for cost in like a decentralized compute Paradigm so um yeah um probably like the most uh

interesting thing that's going to occur here is uh in AI inference um so you know fvm will be um well equipped for this because they've you know um subsidized mining to such a huge extent that there's been a massive infrastructure build out out that has very high levels of like compute on GPU and and CPU and and RAM um in data centers all over so you know the miners on that Network even though currently um you know it's uh you know a lot of the miners are kind of winding down or growing dormant because the subsidies for mining are decreasing uh there could be a lot that's just waiting for the use cases in like AI inference for example to come back online and then the

infrastructure will be there um which is pretty cool uh um AO is taking more of like a organic open model where you know people will join the network um without massive up subsidies and um hopefully like it'll be driven by you know um you know just demand in the marketplace as it has been so far um yeah historically um you know inflation and like levels of subsidies for miners on filecoin has been pretty high it's lowering a lot uh but consistently throughout for our weave it's been it's been low so there's been like less uh build out but um you know it's always just kind of kept up with the demand for for storage um yeah and then uh the storage offering convergence you know we

discussed this already I think it's going to be pretty fascinating um because of the implications that that each have but um yeah so key summaries are um yeah you got the actor model native storage access um and the and main thing is the the convergence um and for differences is you know um yeah fvm focusing on being a hypervisor and a and a normal kind of evm blockchain first implementing WM support later um ao's more General system image that you can load up and boot whatever kind of uh process you want inside of and then pass messages around to other nodes um so right now um

yeah that's that's what that's looking like um and then the the kind of how that's evolved over the past like month or so is um you know um yeah mainly uh like I I would I would I would have been a lot more kind of unbiased like a month ago I would say um you know saying that like let's see what happens with these two but in the past month the amount of like developer attention on AO has been um like massive I think because uh the kind of uh AO team are building you know way more quickly and way more perhaps scrappily just kind of like putting stuff out there iterating fast not being too worried about um the the implications because it's just a test

net there's less to worry about naturally um so that's that's allowed developers to experiment a lot faster and attracted people that are building like you know autonomous AI like agentic systems on top of it you know starting new defi Primitives and um you know building like uh you know stable coins and dexes and all the kind of stuff that you would need um uh from scratch as well instead of like reporting it over from evm um so it's it's interesting to see like all the opportunities that emerge through like a more wasm native uh virtual machine um yeah and then uh and then because uh fvm is in main net already and um I think in general the team that are much more cautious about uh rolling out new features and so on it

just it just evolves more slowly um you know uh obviously it super important to make things secure um so it's probably smart of them to do that as well um but yeah it's uh it's just really exciting times right now kind of in the in the AO ecosystem so I could encourage developers who are interested in like building a new um compute paradigms to uh yeah build that out um but yeah that's uh that's all I've got for you guys so thank you and um yeah let me know if you got any questions [Applause] hey hello um thanks for the presentation

that was that was good um I wanted to ask so if there's all these def five Primitives starting to happen on AO say MH uh you see in the future um the like the liquidity also like as as it is now between l2s is it also going to happen you think it's going to happen also with AO there's like even liquidity going into that which split with all these other chains yeah for sure there will be Bridges and like you you'll see Integrations with really popular bridges onto you know any network where there's opportunity for sure so um yeah there will be like liquidity moving between

like AO and other l2s and so on for sure but having a natively implemented you know um like stable coin for example is also so nice so until like tether and you know circle like usdc come and say like okay we're going to deposit like deploy our own smart contracts natively and natively support AO for example or whatever other network it might be um it's really important to have yeah like a like a native stable coin that's not dependent on some third- party bridge that not everybody would necessarily want to trust so yeah yeah so are people rebuilding The def5 Primitives in wasen now on on the AO

chain um yeah yeah um yeah it's uh definitely a lot of work but uh yeah it's uh yeah and people are also building like a evm compatible chain that sits on AO as well and and doing that so yeah both are happening yeah okay what are you more I mean bullish on like what do you think makes more sense because like Reinventing all the D5 Primitives and wasm doesn't sound like very like do you want to repeat all the Bucks that like the evm had at the beginning stuff like that yeah yeah that'll be a lot of extra work for sure um but uh in the short term for sure yeah just like copy pasting you know things that are already audited you know

definitely makes more sense um but in the long term I think yeah having stuff like natively in wasum on AO will be very interesting like there'll be just more um like things you can do you know it's like less limited as a as a environment so yeah both are necessary well I guess you're doubting the necessary word but but I think yeah it's uh it's fun to see question yeah I super late can you give us 10c can you give us a 10c tldr of AO

versus 10 seconds can you put the slide back up um yeah okay 10 10c difference is uh yeah um okay we got we got mainnet on on uh file like fvm that's been around for a while they're focused mostly on building an evm uh inside of it first um which is what you can use and and enjoy today like staking fill and you know lending it to minors and so on um and AO are focused on WM first so yeah I think that's the most important important tldr difference yeah yeah cool all right thanks everybody

[Applause]