Ep.54 [EN]: Chaofan: Scaling Solana with Solayer's Infini SVM
37:03 Share

You are now listening to Hotlong. I'm the host, Mabel Zhang. This show attempts to shed light on the broader Web3 community. The guests will include Web3 operators, investors, or ecosystem participants of other forms. The show will also have a special focus on the Asia-based Web3 world. Connect East and West. Enjoy the warmest Web3 conversation.

Welcome back to the latest episode of Hotmong. Today we have the co-founder of Solayer, Chao Fan, to be here with us. Do you want to say hi? Hi. We'd love to get through your background or your journey founding Solayer. So I'm a core engineer at Solayer, mainly building the underlying stacks for the Infini SVM chain that we're going to publish in this year.

Previously, I was a PhD student at UC Berkeley, but I later dropped out. And I also co-founded Fuzzland, which is a security company. We have rescued over $30 million from hackers and we also do market making and arbitrage. And then how did that lead to Solair? So it's actually a quite fun story. So Jason has another company. Jason is the CTO of Solair.

He has another company called MPC Vault. MPC Vault was working with Fuzzland to do some security related stuff. And we were having a launch. And Jason gave me a pitch when Seoul was like $120. His pitch is basically just saying like why SVM is going to be better than EVM. Why that we need to have a better EVM.

like a faster chain on SVM. Shortly after that, what he says converted to be true, so it gets really high. The price gets skyrocketed and also Solana gets really congested. At that time, I was really convinced by his pitch and we decided that we should merge together and he later acquired us.

But then like MPC Vault, how did that turn into Solair? Like it's the company before Solair basically. Right, right. How large is the team right now? So just for engineers, there are five and there are a few others, designers, BDs and also operations guys. So yeah, you mentioned a bunch of things related to hacking. I think we can go back to that later on.

So I was curious, is it possible to run the same thing on Ethereum with Solair's approach? I think it's not very likely because state access is deterministic on SVM, but for EVM, it's not deterministic. Basically, when you send a transaction on SVM, the blockchain knows what state you're going to access. But for EVM, the blockchain does not know until it executes it.

and the technique we use is based on the prediction of the states. Adding also the state access prediction onto it is going to lead to a very significant overhead and leading to too much randomness. Although it's still doable, but it would significantly downgrade our technique's performance and it's also very hard to be implemented.

The first fundamental question I would like to ask, just for the sake of audience to know, could you compare the technical approach of EigenLayer with SolLayer? So SolLayer started as a restaking protocol, and there are actually two parts of the restaking protocol. One is the Axel AVS, which is the same as EigenLayer, providing the insurance for all those participants.

And there is another part which is called IndoAVS, which is that we use the stake cell to accelerate transaction landing and also securing the block space. The benefitter or the users of this IndoAVS are mostly DeFi protocols and also market makers.

Basically, it's a plus version of EigenLayer. When was SolLayer as a restaking protocol, this idea come around? I think it's around June last year. It's after EigenLayer. You did start with restaking, but then in your 2025 roadmap...

Why did you decide to expand from a restaking protocol to a hardware-accelerated SVM chain? So this is mostly based on what we see from our IndoAVS. The first thing is that Solana right now is very congested. It's very hard to land a transaction if you don't have a priority fee or you bribe those validators.

And this gets even worse when there was like a Melania launch, a token launch. No one can buy into that token other than the bots. And so we think that we need to actually have a sidechain that could be much faster than Solana while helping Solana process some part of some set of the transactions that are originally sent to Solana.

So this is the underlying thoughts that we are going to build our SVM chain, which is called Infinity SVM. Interesting that you didn't use L1 or L2, but instead you said sidechain.

So I guess like what other blockchains would you compare yourself with? Well, from a technical standpoint, not marketing or promotional standpoint. Do you even consider yourself an L1 or L2? Does that even matter? Actually, I think we're between L1 and L2. And the most similar blockchain that is to us is Monad.

So the underlying thought is that although we have our own consensus mechanism, we don't rely on an L1 to do the block production and verification. We are more like having a POA set of validators who are going to get all those expensive hardware and processing the chain at very fast speed. And there will be also a set of verifiers.

who are going to verify the transactions that are produced by the validators in the PoA set. So those verifiers can be participated by anyone and anyone can just run the verifiers and get a chunk of the whole chain. So the chain will be split into different parts and

The verifier's task is to re-execute some part of the transaction sent to the chain and see whether the execution result matches with what is produced by the validators. If there is some bugs in the validator or validator is trying to be malicious, doing some censorship things, then the verifiers could identify them. But the reason that I don't say that it's an L1 is that because we still rely on Solana.

So whenever the censorship happens or whenever the POA set of validators are all done, then we fall back to Solana to do different kinds of actions like re-election and also forced exits for all the user forms. So it's more like somewhere between L1 and L2.

And regarding a comparison with, for example, Monad and MegaEast, I think our consensus mechanism is pretty much different from them. It's similar to Monad somehow, but the underlying technique is pretty much different. Can you expand a little bit more on the comparison and contrast with Monad? Because you actually brought it up.

It's interesting to hear from an SVM chain that they're saying that their technical approach is similar to Monad. Yeah, so Monad is actually an L1 and they have a fully working consensus mechanism that does not rely on any other chain.

The way that we're similar to Monad is that we are using a similar approach, which is that there will be a leader who is going to process all the transactions. The results are going to be verified by the rest of the network.

But the way that we're different is that our leaders is a very small set instead of all the participants in the network could be the validators. And in the meantime, we also, because like we're having very high TPS, we do not like have every participant in the network to re-execute all the transactions in our chain, but just only a small chunk of it.

So like would each of these small subset, I should say, be working on different tasks or the same group of transactions? They will work on different tasks, not the same task. Okay, so there won't be any like situation where there's a conflict. So the only situation that there will be a problem or like there's like...

lack of consensus, like when it doesn't really align with the Solana consensus? What would happen if that happens? Yeah, so there are two participants in the network. One is the validators who are producing. So it's like Solana's validator who are producing all the results. And there is another set of more generic role, which is called verifiers, who are validating the networks. Those transaction results produced by

validators and those verifiers could fall back to Solana to re-elect a new validator.

Ah, that's how it works. Understood. Before we go into the end user experience, I wanted to ask what's the participation threshold for both Verify and Validator? So for validators, right now we're going to make it into a very small set for only folks that we know because we need to send them hardware to run those validators.

But in the near future, I think we're going to make the hardware much cheaper and we're also going to open source the PCB and all the circuit designs and everyone could just build their own hardware and run it.

And at that time, everyone can be a validator. And for verifiers, as long as you stick as so, we haven't decided a threshold yet. But basically, there could be hundreds or thousands of verifiers in the network. I see. And then right now, who's producing these hardware's?

Right now, we're just using existing like commodity hardware. So like FPGA boards and we're writing logics onto it. Yeah, but ultimately it's going to be like a PCB board, not like a FPGA board. Whoever's running these hardware, I would assume that they're likely the ones who are running the Solana validators themselves. Otherwise, like it's going to be tough to run it themselves. Yeah, that's what I figured, understood. I think Anatoly had a similar idea of like,

gradually the hardware threshold is going to be lower and lower and it's easier and easier for someone to run a validator or even for Solana back then. For any of the end user, when I said end user, I meant consumers, when they're trying to interact with Solair, what would that experience look like?

So we're designing two kinds of experience. The first experience is that the user could directly send a transaction on Solana and interact with any Dapp on the on the Infinii SVM chain or even bridge in a single transaction and bridge it back.

So, users can basically sign one transaction, send their assets to our Infinii SVM chain, and do all those trading-related things, and the settlement could all happen in a single transaction. And later, the funds will be bridging back, and all these could happen in less than one second. And users can barely feel it, but they could access the liquidity on our chain and also on Solana's chain.

So this is going to be one kind of experience. And of course, we are going to have users who are using the native Infinity SVM, not on the Solana chain, but just on the Infinity SVM. So the experience is going to be pretty much similar. Right now, we find out a way to just let Fantom and Soul Flare wallets to directly sign for the transaction for our chain. But they don't need to like using MetaMask

you need to like switch the RPC every time you want to switch a chain. But for us on Phantom, you don't need to do anything. You just need to click sign the transaction and your transaction will be signed and broadcasted onto our chain. And the work was done primarily from Phantom's side or from your side? So it's going to be an SDK that the app can integrate. And by using the SDK, it's going to connect the Phantom or software wallets

and make it to sign a unique type of transactions. And that transaction finally got routed onto our chain. I see. So you do also work on these like Dev SDK. Sorry, SDK for the... It's like a Dev tooling, technically. Okay. Right, right. I think it would be interesting for you to talk about like, well, you guys stated 1 million TPS on the... I think, is it a white paper or roadmap? Yeah.

Can you maybe talk about how Sololayer pushes that 1 million TPS state sync efficiently to thousands of nodes? We're using a similar approach as Solana, which is called Turbine. It's like the current leader pushed the state sync to a set of nodes that are close to it.

and then those close to its nodes would then broadcast the transaction to other nodes that are close to them. And this keeps going like a tree structure until it reaches the very end users.

By doing so, it's like the validators just need, or the current leaders just need to connect to very few nodes and just use very few bandwidths to push out the stay sync it currently have. And those stay sync would then be propagated by the tree of the nodes. And it's not going to be like

Ethereum where the nodes have to do P2P and be like a really undirected network, which the nodes have to connect to a few other nodes and this keeps going and a lot of bandwidth is wasted by just nodes talking to each other about things that they already know.

And we've actually just added a few more modifications to the turbine approach. We're adding things like sharding. So instead of just one node propagating to a few other nodes, it's like a cluster propagating to a few clusters with different shards.

That's like gossip. And another thing we do that is really new is that we find that some of the RPC providers, they don't need to have very low latency. They could allow for like one or two second latency. And bandwidth is actually very, very expensive. If we are going to push 1 million TPS, it's going to use more than 10 Gbps network link. And it's very expensive.

So we also use like a PCDN service. Those PCDN are like, so there are a lot of Web2 folks who are interested in contributing their bandwidth for very, very little money. And we just use those kinds of folks' bandwidth and serve the normal RPC users traffic through those Web2 folks.

So the starting point of you guys running a hardware accelerated SVM chain is that you guys also think Solana needs to be scaled, even if they are already considered a scalability chain. I think right now Solana cannot really get to its real performance because it's limited by the slowest node in the network.

The slowest node is not giving the vote for a very large block that are produced by more performant validators. This is making the block size of Solana to be really small, but actually it can be much larger. One thing we try to resolve is that we are removing those slower type of nodes in our network by just having a PLA set of validators.

And the verifiers could be slow nodes, but they will not limit the network. Yeah, that's the overall thinking. And what kind of use cases would be particularly benefited from the design or the performance of Solair? The most simple use case would be D-Ping. So right now D-Ping are not propagating all their messages on Solana. Instead, they're having other side channels to propagate all those messages.

The reason that they don't use Solana to propagate those messages is because Solana is relatively expensive whenever folks want to send transactions that could really land. Or sometimes it could simply not land when the whole chain is really congested. For us, sending transactions is going to be very cheap and transactions are not going to be dropped.

I think Deepin would be a good use case for our chain and they can use our chain as a channel to synchronize with other devices. Devices could propagate messages to where the message should be received. So when you said the transactions are not going to be dropped, it kind of just reminded me that users are often sandwiched while trading on Solana. They lost millions of dollars.

Like how specifically you're addressing this problem then? I think this is actually a really easy to solve problem on Solana. The Solana Foundation just could just make a rule saying that validators who is doing sandwiching will be slashed, then it will be solved. Validators are not going to sandwich if they're going to lose their hundreds of millions of dollars stake.

And I think we're going to make this as a standard rule on our chain. And whenever a PLA set of the validators in our PLA set get conduct sandwiched, then we will immediately kick them out. Those verifiers would be the sentry who are noticing and identifying the potential sandwiching behaviors, and they will vote to remove the validator from the set.

That's a very interesting trade-off. So like, I guess from your standpoint, you think savaging is more of a malicious behavior regardless, even if...

I think some people would consider MEV as like a healthy signal of that chain. MEV itself is the healthy signal because you need to have liquidities to be rebalanced whenever there are significant price moves. And this is commonly conducted by arbitrage. And arbitrage, we believe it's healthy and we actually have the

have built a lot of mechanisms to further facilitate backrunning, arbitrage and liquidation for our chain by providing like a hook environment. But back to Sandwich, Sandwich itself is not making the price more rebalanced. It's just like extracting from the users. And Sandwiching itself should not be considered as like a healthy product.

MEV behavior. Are there a lot of large, well I shouldn't say large, are there a lot of validators on Solana that are actively doing this? Yeah, there is actually one very large validator actively doing this and there has been like a rumor saying that there is a dark pool of validators, RPC providers, and a few other participants just sharing those private transactions or transactions that are not supposed to be public to the sandwich bots.

Oh, that's crazy. So people are running private RPC, but then they're not technically protected. Right. Wow. Okay. Well, I mean, on the one hand, obviously, like, you know, having such complicated, like market behavior, probably signals that Solana has a healthy network, but that's a little bit too much, definitely. Like one thing I know about BNB chain is that almost all the validators sandwich the users. I thought it's...

going to be slightly better on Solana but I guess like there's still such cases. I want to pivot a little bit. I think somewhere was mentioned that Solair may use AI technology in accelerating the chain. Can you maybe expand a bit more on it? Yeah so it's not fancy AI, not LLM or like fancy AI models. It's more on the statistics side. We use statistic models to predict the states.

So a simple example would be like there are 100 main coin trades. Each trade would depend on the results on previous trade. Otherwise, you're not trading correctly. Then because of this, it's not possible to handle all those transactions or trades in parallel for a chain. The chain has to execute the trades one by one.

because they depend on the previous state of the trade. And because of this, a lot of chains cannot, like fundamentally all the chains right now cannot achieve parallelism on conflicting workloads or this kind of meme coin trades.

But instead of just stopping there, we use a mechanism called statistic prediction. So we predict the potential state of previous trades and we use those predictions to do the pre-execution and use those pre-executions to facilitate the overall production of the block.

So yeah, and the sequencer will decide what is the pre-execution result they want to use because there could be hundreds of predictions for a single transaction and there could be hundreds of results and the sequencer will decide which one, the optimal scheduling they want to use to best use all those pre-execution results.

Have you seen any other peers doing similar things, but at the production stage? I know Monad is doing similar thing, but what they do is more on the primitive side. I don't have access to their code, but they claim to do it. But the performance of their chain is not so good. If they could really achieve it, then they could get 1 million TPS rather than 10k TPS. Yeah.

Yeah, you earlier mentioned about staking as SOL. So I wanted to raise a question related to SUSD and SOL. How does the roadmap for SOL layer SVM chain tie to the issuance and circulation and maybe usage adoption for SUSD and SOL? So we're using SOL as the underlying assets of our chain. It's going to be the interest bearing SOL as the gas token on our chain.

and SUSD is going to be the native stablecoin and shortly we will also provide a Yelp-bearing Bitcoin called SBTC. Are you working with Zeus or are you doing it yourself? We're doing it ourselves. What role does Layer play in the whole ecosystem then? So Layer is going to be part of the consensus mechanism. So to join the proof-of-stake set, the user has to either stake as a ZO or stake Layer.

and the layer would also be for governance. Ah, okay. So when you said earlier, when you said validators, like they can either do as sole or layer staking. Right. Isn't that going to cause like conflict of interest? As in like people would prefer just using sole instead of layer and would just create lack of demand. No, it wouldn't because as sole is going to be like...

I don't want to say it, but as so is less fluctuating than layer. And layer is going to be overall getting more yield from the emission. So there is like, I guess, a pool for you to provide yield for both sides. And then obviously layer, given its volatility nature, it's getting more yield. That's basically what you're saying. Yeah. And there will be a slow emission of layer. Yeah.

there will be new layer minted to generate the Yelp. Right. I've never seen such validator mechanism having like two tokens taking for like validator requirements. So I thought that was quite an interesting thing

or innovative approach. Yeah, it's like restaking, like you're using valuable assets as an underlying way to ensure the security of the chain. I'm less familiar with EigenLayer, but did EigenLayer also did the same thing? No, they're not using their own token.

But yeah, they use BTC and Ethereum to do. Right. That's why I'm saying I thought this is quite interesting to watch. Well, I'd be keen to follow up in 6-12 months and see how this plays out. I want to switch gears a little bit. CMD 2.0a just passed. What I meant by "passed" is that this whole process already passed. It failed.

So what's your personal take on SIMD228, given that I think you probably have a lot of insight about Solana Validator? My take could be inaccurate. It doesn't matter. It's just your view. My personal view is that SIMD228 is going to reduce those rewards for some smaller validators who are actually contributing to the security and decentralization of Solana.

But after reducing their rewards, they become unprofitable or even lose some money as they need to spend to do the voting. And because of this, they are going to be forced to shut down. And stakers are going to move from those smaller validators to all the bigger validators who are still profitable. And Solana would get reduced amount of validators, reduced amount of...

continent it's on and overall it's getting more centralized. And another reason that we voted no, it's not due to my opinion but due to the opinion of our stakers. Most of our stakers after we do a weighted sampling say that they want to vote no.

Oh, you actually questionnaire them. Not many validators does that. Right, we actually talked to a few of our stakers, but there are also like stakers we have no idea who they are, so it's not possible to reach them. Right, that's quite interesting. I mean, I always know that this is going to lead to a less decentralization of the network,

But when you said the smaller validators are going to be less profitable, it's because of the marginal cost. When you're at larger scale, you are technically like you just spread it thin the cost of running it. That's why like the bigger validators are less threatened by so. Is that what you meant? Yeah, that is correct.

Okay, understood. But then you guys also decided to sell 10% of your SIMD 2 to 8 vote on Meteora. Why is that? So one of our very large stakers said that they don't want to vote on this and they don't really care about those SIMDs. So we decided to sell a very small portion, which is smaller than what they stake, to the Meteora pool. And

distribute the reward back to the or the amount we sold back to the to the stakers actually a majority of the same d228 is not sold on meteor but through an otc deal with uh with sole place but then like the uh proceeds of that sales are distributed to all the stakers not just that one yes i thought that was quite interesting perhaps like

layer can be also part of this like governance process in the future now that i know that you guys uh went through a merge you guys have been doing quite a lot of like public good works um that may not necessarily you know benefit the protocol or the chain directly um for example i think you guys try to save the fund that the one inch market maker got hacked but obviously you weren't fast enough so like it still didn't

really, really help. But then you guys did try MVV some other cases that you try to publish the details for the security incidents. Why are you doing so? So this is mostly done for Firstland. We are like an active incident response team and we try to rescue funds from every attack, not just for our customers, but also like we detect potential threats and try to stop those threats.

But then like, do people care if you spend, you know, time on things that are not so rare? I guess like it works pretty, this strategy works pretty well for first land. We get a lot of customers who were previously hacked and with some funds rescued by us. So I guess, yeah, I probably should not say that. But overall, like, yeah,

We get a lot of customers from those incident response tweets and also from rescuing deer farms. Right, so it's also like a good marketing strategy approach. I see. And then what are some of the...

protocols that you guys recently hacked? I think the most recent one is Moby. It's uh, I don't know what that protocol is, but we rescued around 1.2 million dollars. Oh, I see. But then you also hacked Timed Off One, I think? Yeah, but it's a small one. But we don't call it hack, we call it like rescue. Like white hat hacking, I guess that's what I'm trying to say. Right, right.

I would like to learn a little bit more about the team culture. It does sound like you guys, the reason why I was asking this is because like, I think you guys are all quite engineering focused, but at the same time, both yourself, I think Jason, the CTO, were quite vocal and then are very good at speaking about, you know, what you're doing. You know, people call it public building or build in public. So we'd love to learn about the team culture.

Right, so currently we have around five engineers and we're really collaborative but we don't really talk that much because all of us are... I guess I'm not going to talk about me but

all my colleagues are really smart and they don't need to spend time on new stuff and they could just learn it instantly somehow. And yeah, we have a pretty good team culture. Everyone is helping each other and we do CTF competitions, we do hacking togethers. So we actually

uh, like, uh, work really well with each other. Yourself, I think a lot of people who are not part of the Solana ecosystem knew about you because you kind of just, like, took deep investigation on the two million, um, got, how should I even put it? Like, it's not even hacked. It's just, like, I guess, scammed. Um, yeah, I rocked.

Maybe you can share a little bit more about your deep thinking in terms of why you decided to go into that investigation and then what were the results? I think the overall reason is pretty much simple. I lose 2 million in this coin and

And it was actually pretty easy to find them out. They're using FixedFloat, which is a centralized exchange. And we have a technique to trace the font through FixedFloat.

It was actually pretty funny, I was just tracing it for fun but later linked it back to a lot of other tokens like Melania and also M3M3 and I decided to make it public.

But then like in that process, do you think the company or I should say the protocol also get a lot more exposure? Yeah, I guess so. Like at that day, a lot of folks start to buy a layer.

Oh really? I didn't know that was the case. I see. Yeah, we didn't rock them to be honest. And every one of them, if they hold till today, they're like getting their... What would they have doubled? And then I saw you were hiring a security researcher.

1 million a year, is that real? It was real. In the last bull market, we're actually making a lot more than that. So it's actually not uncommon for market-making companies or security companies to give 1 million every year to their engineers. But right now, we're in bear market, and it's actually really hard to get that much profit and profit.

and paid like 1 million every year. And to be honest, like after I sent that tweet, a lot of folks approached. But like what I tell them is that like giving out 1 million every year is not because I'm rich enough, I'm wasting my money. But it's more like if you're worth 1 million, then I would like to hire you with 1 million.

So what next? What's the upcoming like six to 12 month roadmap for Soul Layer Chain?

Right, so right now we already have our internal DevNet up and our partners are actively testing their projects on our internal DevNet. And maybe in next month or Q2, we are going to release our testnet and users could start playing. And in Q3 or Q4, we'll have our mainnet. Do you have any, I guess, decentralized applications line up ready?

Yeah, but I cannot say who they are. Are you guys going, like hosting any hackathons to attract builders? Or how do you get people to build on Solaire?

No plan yet. We're getting a lot more existing projects who have really mature code and ecosystem. Our initial plan was to attract most of them and also attract all those market makers to provide very good liquidity on our chain and very good trading experience. But

I think we're going to host some events to attract the users, but not really just for the developers. And then for the market makers, is that going to be incentivized? Yeah. Well, it was fun chatting, I think, after this conversation.

a lot of our audience would have much better idea and then the full picture of what you guys are building. I think you guys have a very interesting and I wouldn't even say like peculiar approach of doing marketing, but obviously it does work in like today's kind of market. People are kind of trying to grab attention from each other. So good luck with everything and thanks for joining the Hot Bomb. Thanks.