Anatoly joins us to discuss why Solana is uniquely designed to utilize ever-increasing bandwidth and compute. We cover Solana's core thesis, how blockchains benefit from hardware improvements, the fundamental flaws of L2s, Anatoly's favorite applications, decentralization and more! Anatoly is a legend, you don't want to miss this one!

- -
Follow Anatoly: https://twitter.com/aeyakovenko
Follow Mert: https://twitter.com/0xMert_
Follow Garrett: https://twitter.com/GarrettHarper_
Follow Lightspeed: https://twitter.com/Lightspeedpodhq

Subscribe on YouTube: https://bit.ly/43o3Syk
Subscribe on Apple: https://apple.co/3OhiXgV
Subscribe on Spotify: https://spoti.fi/3OkF7PD

Get top market insights and the latest in crypto news. Subscribe to Blockworks Daily Newsletter: https://blockworks.co/newsletter/
- -
Use code EMPIRE30 to get 30% off Permissionless 2023 in Austin: https://blockworks.co/event/permissionless-2023
- -
Timestamps:
00:00 Solana’s Core Thesis
10:15 The Scaling Bottleneck
13:33 Do All Blockchains Equally Benefit From Hardware Improvements?
17:32 Pros and Cons of Parallel Execution
20:47 The Realities of Decentralization
24:18 Solana’s Design Moat
27:45 Permissionless ad 
29:11 The Fundamental Tradeoffs of L2s
43:46 How Should Solana Approach Business Development?
47:47 The Role of Solana Labs and the Solana Foundation
49:22 The Compressed NFT Opportunity
53:06 We Need to Fix DAO Governance
54:48 Anatoly’s Metric of Success for Solana
57:33 Why Anatoly Hasn’t Left Crypto
59:38 Why an Average Joe Would Use Solana
01:01:14 Rapid Fire
  
- -
Resources

Endgame: Proof of Governance by Jon Charbonneau
https://www.youtube.com/watch?v=toPd1vgHjVE&t=0s

- -
Disclaimer: Nothing said on Empire is a recommendation to buy or sell securities or tokens. This podcast is for informational purposes only, and any views expressed by anyone on the show are solely our opinions, not financial advice. Santiago, Jason, and our guests may hold positions in the companies, funds, or projects discussed.

If we're limited by bandwidth right now, in 2 years, there's going to be twice as much.

Maybe we'll need twice as many cores to drive it because there's twice as much data to sign and cryptographically verify.

But these are all Moore's law problems that just kind of solve themselves.

But what we need to make sure in the design is that we can always increase the bandwidth.

Like if the network capacities go to 100 gigabit in like 4 or 5 years at every data center, they kind of are in a bunch of them already, then Solana could run at 100 gigabit.

Like we shouldn't be limited by the design, we should be limited by the hardware.

If you kind of design around that, you have a very, very scalable system.

It's very elastic to demand up to that limit of hardware.

And I think we've done a pretty good job with the design because what I see is that the Firedancer folks, without having to make any design changes, are showing much higher throughput by just doing straightforward, okay, you guys had to ship this thing as fast as you could, let us rewrite it in a better, more scalable kind of just raw software architecture way.

And they can really demonstrate that all the components can scale very elastically with hardware.

Welcome to another episode of Lightspeed.

Today, we are joined by Anatoly from Solana.

Yeah, yeah, We are pumped to have you on.

I've been studying more and more about Solana and every time I do, I get more excited.

So I wanna start high level to set the scene and really frame the conversation.

So when looking at all the different design choices in crypto, it's easy to really get caught up on what you're building or what's being built and not why it's being built.

So I want to start with the end with Solana.

And I think that will then lead to why you went with an integrated global state machine and not a chain that relies on charting state with roll-ups, but instead relies on like ever increasing bandwidth and compute.

So the, what I thought was really cool about crypto was that it enabled permissionless markets.

And if you've ever traded in any permission markets, like I traded like as a total amateur.

I had connections to interactive brokers and a bunch of other places and I've run my algorithms.

Every time I thought that I had alpha, my data would arrive a little slower and my orders would be submitted a little slower and just felt like everybody was always getting ahead of me.

And that's because those are not fair markets.

And what makes them unfair is that there is centralized actors that get the information before you, they get to submit the information ahead of you.

And that creates massive amounts of value capture for them and losses for somebody an amateur like me.

So what I think is really cool about permissionless systems is that they can create fairness.

And The whole purpose of Solana that I thought was really, at least it would be valuable to the world, is that we can create fair markets at the core information level, meaning that bits are transmitted everywhere around the world as equally and fairly to everybody concurrently at the same time as physics allow.

And that submission is also as fair and competitive as physics allow.

So those are kind of the 2 problems that we set out to solve.

And I think this is very different than most of, probably every other crypto project that started out at that time, because they started from either the store of value narrative of Bitcoin, we want to build a decentralized money, or settlement guarantees.

We want to build a decentralized settlement layer.

And none of those really care about when the information arrives, who gets to submit it.

Like all those problems are secondary and they're interesting problems.

But I think what most people don't realize is if you build the first, if you build this perfect information system, settlement is free.

So solving this much harder problem actually solves all the other ones.

So that's really like, I think, kind of our goal, Solana's vision is to be that like perfect single state machine that everyone can rely on for fair access to the data and kind of see it as a source of truth for the world.

So obviously the information sinking globally at the speed of light is kind of the big, let's say KPI that you're optimizing for.

It's not obviously KPI, but so, and with that said, the first pitch for Salon, the elevator pitch, the catchy slogan was NASDAQ at the speed of light.

Is that still, you think, how people should talk about Solana?

I think there is like kind of a bit of a trap there to think of it only in terms of trading.

This is something that I realized as building this is that the building a NASDAQ competitor is, and going out there and trying to get stocks listed, I think that could be a thing that people do, but I think that's a very small part of Web3.

I think the fact that you have open permissionless markets really disrupts how the web is monetized, how people coordinate globally, and all the kind of really cool web 3 things that you want to do, that I think are much more revolutionary than reducing the amount of take that Citadel and NASDAQ has.

They're just much higher impact than that.

If everybody in the world has a cryptographic key and they know what to do with it, they know how to sign and coordinate in the DAO, that's effectively full global democracy, full global direct democracy.

They can all deposit into a DAO that says, we're going to buy every coal plant on the planet and shut them down.

If people know that that will actually happen, I think they will actually throw a thousand bucks in there.

And that very, very quickly can become a very meaningful change that is enacted through this.

So I think crypto is just much bigger than markets.

But what makes crypto interesting, I think, is the fact that people have cryptography.

This is, I think Vitalik had a quote that I think cryptography is the first time that defense has an asymmetric advantage over offense.

It's not just like you cannot break it, but an atomic bomb you cannot break at 2519 signature.

This is kind of like the really cool thing about it.

It will take infinite energy to go break that key.

And I think That's a very empowering thing.

And I honestly think that if we could give humans the cryptographic keys, it doesn't really...

Like the blockchains, all the stuff, I think those are implementation details to a large degree.

The way that I'm interested in attacking the coordination problem is through the thing that I'm good at, which is like low latency optimizations, making things fast.

And I think a chain that focuses on that, on synchronizing state, the least amount of overhead as possible has a place in that world where everyone has cryptographic keys and they're doing stuff on chain with them.

So my second question, and you kind of answered it as well, is, so you mentioned that the other chains maybe start from a place of like settlement, which I believe you think is the most boring part of finance, store of value like Bitcoin, the sovereign aspect, but then yours is globally shared state, which is actually a pretty abstract concept.

I know some people who aren't too familiar with maybe what that means.

Like, how do I sell it to like somebody that's like, this is a digital gold.

So that leads me to my question, which is how do you rally people around globally shared state?

What makes you think that's the right direction to pursue?

Well, I know that it eliminates ways for this arbitrage, I think parasitic value capture.

When the system gets faster, lower latency, more competitive in terms of who can write.

Like the amount of opportunities for somebody to extract unfair value goes away.

And that's really important to like all the DeFi nerds.

I think the reason why Solana has like a resilient DeFi community that's been able to take these major hits is because they get that vision, that they know that if I build a project here, I'm not going to get screwed by all these intermediaries that can extract massive amounts of math.

Like on Ethereum, because it's like, there's like this idea that fees are good, but that's actually value that's extracted from the user.

There's 0 chance that Senator Warren of the SEC would allow NASDAQ to take the kind of fees that Ethereum does.

They would shut that down like instantly.

Like all those regulators have like created everything around finance in the United States to prevent that kind of fee extraction from occurring.

So I think for cryptos to be competitive, it has to be better than the existing regulatory systems, not just better because it's open and permissionless, but actually better.

It's got to be cheaper, faster, more open, And then I think it can kind of like start replacing those systems and like be a value to the world, you know?

Let's talk about some of the ways that Slana is actually going about doing that.

You said that the final bottleneck is DA bandwidth.

What do you mean by that and how does it apply to how performant Slana is trying to be?

So like, this was an intuition that I had early on.

We didn't even call it data availability bandwidth.

It was really just this idea that the chain, you have a quorum, right?

This network of nodes that all have to synchronize the state.

That's the current active set of nodes in the quorum.

The fastest they can synchronize that state is going to be limited by the slowest 1 third of the network, your Byzantine threshold, whatever, however many need you need to get to consensus.

That slowest 1 third is going to limit the throughput.

So you have to basically optimize and increase that throughput.

And what's cool is that when you increase that throughput, you can increase the quorum, because what's very, very trivial to build, A very dumb consensus thing you can build is that every node votes and propagates that vote to everyone else as just part of the way the chain normally schedules all the transactions and all the events.

This is why Solana consensus is implemented as a native Solana program, like a smart contract.

And that's why votes are normal transactions, is because when you increase TPS and capacity, and therefore the data availability bandwidth at each node, you can actually just add more nodes and increase your quorum.

And therefore, that number of users that are getting the data that's around the world increases and you have a fairer and more open system.

It becomes harder for you to delay the data from any part of the world and therefore extract value, Information value from that.

So this is a very, very kind of brute force approach because it's not algorithmic.

It's very much based on if we are limited by bandwidth right now, In 2 years, there's going to be twice as much.

Like if the network capacities go to 100 gigabit in like 4 or 5 years at every data center, there kind of are a bunch of them already, then Solana could run at 100 gigabit.

Like We shouldn't be limited by the design, we should be limited by the hardware.

And if you design around that, you have a very, very scalable system.

It's very elastic to demand up to that limited hardware.

And I think we've done a pretty good job with the design, because what I see is that the Firedancer folks, without having to make any design changes, are showing much higher throughput by just doing straightforward, okay, you guys had to ship this as fast as you could, let us rewrite it in a better, more scalable kind of just raw software architecture way and and they can really demonstrate that all the components can scale very elastically with hardware.

I'm sure you saw this recently on X.com that, you know, people think that all blockchains and obviously all computers do benefit from hardware advancements and bandwidth advancements and whatnot.

But you know, what I said was that Solana benefits, you know, like the benefit that you get from each chain is not proportional, like they're not the same thing.

Can you describe, like, how you think about that?

Like, if when the hardware increases, the improvements are made, do all blockchains benefit the same?

So like that kind of idea of like, I kind of made a response to your post saying that it shouldn't be that Solana is scaling with hardware, it's that hardware is all you need.

And this really came from my experience at Qualcomm.

I was there from like 2003 until like 2015 or something like that.

And this was when we went from flip phones to smartphones.

And literally every year, we had an architecture change at the CPU and a process change and ARM was getting better.

It's just, it felt like Moore's Law wasn't even going at 18-month interval, it was like a 12-month interval time.

And if you didn't write your software in such a way that it's just going to get better by the next chip, you have to rewrite it.

So it was like a waste of time to build it in a way that doesn't scale well with more hardware.

And that was just really burned into me as I was doing the software development, trying to keep up with like the hardware improvements.

That's kind of like the idea there is that, because hardware can scale at this exponential pace, if you design your software such that it just naturally takes advantage of it, whatever inefficiencies you have in the algorithm, Maybe there's an algorithm that's just way more efficient in some parameter.

So all the time you're wasting trying to optimize for, you know, like all, like squeeze out extra percentages right now, you should just do analysis, figure out where you're not scaling with the number of cores and fix that because literally in 2 years, you're going to have twice as many cores and those other optimizations are going to be half as important.

And this is kind of this exponential curve that humanity's on.

My personal belief is if that ever stops, we should all be working on bunker coin.

We've hit the limit of human civilization.

I personally believe that the number of cores per system per dollar is going to continue to go up because we're not even stacking them in 3 dimensions yet.

These are still single layer, single CPU, not everything.

Some phones do have stack RAM, but this is still super early in terms of how much normal Silicon can be pushed.

So a lot of other chains do get improvements from hardware.

And I don't want to say that, you know, like, I don't think that's what you meant with your quote.

It's just Solana and all of our engineering is specifically tailored around this idea that we need to parallelize and maximize the efficiency across the number of disks, across the number of cores, number of network cards, to make sure that in 2 years when the stuff gets half the price, that everything just elastically gets better.

And if we achieve that, we're kind of like done with the software architecture, we're done.

Like we can kind of rest easy or probably not.

We have to work on all the other stuff in the application layer, which is infinite amount of work.

When you do that, you get to rely on like Intel and AMD and ISP providers with more bandwidth, right?

1 thing that Solana has been known for is parallelizable execution.

Why is it that something like Ethereum didn't originally start with parallel execution?

And Does part of that go into the fact that Solana separates state and programs?

Yeah, so this is, again, from my experience working on embedded systems, like working on DSPs and things like that, if you've ever had to program a DSP or a GPU, or even like an AVX pipeline on Intel, it's a quite different environment.

You have to lay out all the memory and all the lanes for a program ahead of time.

So you have to know all the dependencies of all the memory accesses ahead.

And then you tell the CPU, here's the code that you're going to execute across these lanes of memory, and then it just zips through them.

Because once it knows the locations of all the dependencies, it can prefetch all the memory, thus eliminating any kind of speed hit to a missed memory load or something like that.

Everything just runs through the system as fast as possible.

And that's really translating, like, making the system be as efficient as possible in terms of execution.

But that requires a developer to do this work ahead of time.

They have to know all the dependencies of all the memory locations, and yada, yada, yada.

And that obviously eliminates features like global memory that you can just go and call random memory or jump to a random memory location.

So the way Solana's runtime is designed is from that idea.

The developer does the painful work of figuring out what are all the memory dependencies that the program has and conveys that in the transaction that they constructed.

So then we can schedule all these events and execute everything in parallel.

It's been like, I think, It's been interesting to see how hard it is to do that once you kind of open the Pandora's box and let developers call random locations and do whatever they want with the state.

Because just the way that they compose programs once they're given that option is very different.

You give them the tools, they will like treat everything as a hammer.

And that becomes very painful to rewrite those systems.

So I think it's gonna be very, I think parallelism is going to be really, really hard for EVM.

And it almost, given that the fact that the way EVM is designed, it almost makes sense that roll-ups are the way to go because there you're isolating state with tokens and NFTs through a bridge, and then you have a little EVM environment that runs in that state, that's all global, then you have another little EVM environment, another roll-up that does its own thing.

And that's how you do your state separation.

You've kind of created parallelism that way.

But that's a much more, like kind of, it's far less granular, right?

You can't like compose between these states very easily.

Everything becomes much harder on the composition side.

gonna ask 2 questions which won't be that related.

So I know you said you have like a theory that being able to process multiple things in your head is like improves the brain hardware or whatnot.

First is having said that actually L2s are probably the better way for Ethereum to scale.

I'm curious on your thoughts about Monad, which seeks to do parallel AVM.

And second, so what's, you know, it feels like it should be obvious to most seasoned engineers that you want to scale with hardware.

Yet, I would say you're like 1 of the, I think Solana is probably the first serious attempt at that.

And some people will make the argument that, well, the reason is because you can't run these systems on retail internet, right?

Like at my house, you need kind of very strong internet connections and whatnot so that people can maximize the number of nodes running.

But then you say to that, well, it doesn't matter how low the requirements are, what matters is the incentives to run it.

How come That seems to be an unpopular view.

Well, I think you have to go back to what you believe is kind of like, what is decentralization?

And my belief is that all interesting theories have to be grounded in physical reality.

You have to see actually something physical happen for this to be a worthwhile theory to even discuss.

So, if we're talking about decentralization, how does it manifest itself in the physical world?

And, from my perspective, the key components of decentralization is that the system is permissionless.

That means that I can physically create my own box.

I don't really care about the cost there.

We could argue about whether it's 400 a month is too much or $50 a month is like the required amount for decentralization.

I think that's almost like, if we get to that, if the discussion ends up there, I've already won.

But I think The key part is that the system's permissionless.

And the other part to it is that there's enough people that give a shit that it's running, that even if something fails, it'll continue it.

And that latter part is all about the incentive.

So in a system that is very cheap to run and cheap to participate in, I think it works for Ethereum, but this is not going to work for anyone else.

If you started a very slow, expensive chain now that no one's participating in, You're not going to get enough incentives for people to actually care that it's running, that they would go and continue it.

The fact that Ethereum has succeeded, I think, is because it was early and they created the whole smart contract system and has a lot of very credible, awesome researchers working on it and really working on that towards that goal of settlement.

And it has PMF in what it's doing, but there's no way it's like, if somebody else tried to copy Ethereum and launch it, they're gonna fail for the same reason that if I tried to launch a Bitcoin, like slightly, but slightly bigger blocks, slightly smaller blocks, whatever, right?

Like it's just somebody already captured that part of PMF.

You need to find another incentive structure that will guarantee that people actually care about the state and continue running it.

So Solana is a very performant, technically sound chain.

And I would say it had a lot of updates in 2022 that went under the radar.

And I think a lot of that was with SPF and some network outages.

And some of those were quick and stakeway to QoS, which we can go into that, but it's really just like network stability.

And you had fees, you added priority fee markets, isolated fee markets, which I would like to touch on and how that kind of corresponds to L2s.

You also had things like Solang introduced where you can do Solidity on Solana, you can even do Move now.

I'm curious with all that work that you put behind it, how easy is it for, despite this is me being not technical, how easy is it for another ecosystem to almost just copy all those updates that you've had?

Because to me, Solana is shipping faster than any other chain out there?

And is it really just the community that's the part that's not able to be forked?

It's like, you know, BSD had jails, like I think 10 years before Linux had namespaces.

It takes even this, like those, all those engineers knew each other.

They like, literally we're all friends and we'll get beers together.

It's actually really, really hard to yank stuff from 1 part of the, like 1 operating system to another 1.

So I think some things could be somewhat transferable.

I think you could take SVM and run it as a rollup on Ethereum and encourage people to do that.

I think Eclipse is 1 of the teams that's doing that.

I think we just need more SVM environments out there.

People think, oh, it's competitive with Solana.

It's like more people that give a shit about SVM.

They'll create more value for the entire Solana ecosystem, like Linux.

See all Empire transcripts on Youtube

Solana’s Ultimate Vision | Anatoly Yakovenko