Emin Gün Sirer says the AVA blockchain will scale better than Ethereum 2.0. Others say the problem is more complicated

In the race to build a next-generation smart contract platform, Cornell computer science professor Emin Gün Sirer is convinced that the protocol his startup AVA Labs is developing has an edge – at least when it comes to the complicated technical challenge of blockchain scaling.

The problem has haunted blockchain developers for years. But one morning roughly two years ago, the computer science professor enthusiastically tweeted about what he now sees as part of the solution.

He linked to an anonymously-written white paper that described a “new family” of blockchain consensus protocols that Sirer called a “huge breakthrough.” Soon after, he founded AVA – which has raised $6 million from prominent venture firms like Andreessen Horowitz and Polychain and investors like Balaji Srinivasan and Naval Ravikant, among others.

The goal: to build a brand-new blockchain system, based on the so-called Avalanche consensus protocol. Last month, AVA opened its first public testnet, and the team hopes to launch AVA’s mainnet in July.

Like the planned Ethereum 2.0, AVA’s network will use proof-of-stake to reach consensus. But Sirer says AVA’s approach will allow it to perform faster and more efficiently.

Not everyone thinks the problem is so straight-forward, however.

Ethereum 2.0 killer?

In the past few years, a number of smart contract platforms have arisen, each promising better performance and greater scalability than Ethereum, which, like the Bitcoin network, relies on proof-of-work – an energy-intensive process – to reach consensus.

Each of these platforms has swapped in some variation of proof-of-stake, in which nodes prove their trustworthiness to the rest of the network by locking up (“staking”) some amount of tokens – instead of expending the massive amounts of power required to participate as a miner in a proof-of-work system.

The Ethereum community itself has long acknowledged the shortcomings of proof-of-work, which is why it’s planned next-generation platform, ETH 2.0, will use proof-of-stake, too.

The first wave of “Ethereum killers” aimed at the current version of Ethereum, which remains the world’s most popular smart contract platform. Sirer argues that AVA is designed to outclass ETH 2.0 as well.

According to Sirer, AVA will be able to achieve not only higher throughput (the number of transactions that a network can process per second) but also low latency (the time it takes for nodes to agree that a transaction is valid).

Right now, Ethereum processes around 10 to 20 transactions per second (tps). ETH 2.0 hopes to multiply that number by 64 times. Sirer says that AVA’s testnet already reaches 6500 tps. Meanwhile, whereas today’s Ethereum has a latency of several hundred seconds, and ETH 2.0 is working to get a five-second latency, AVA promises that the network will have a latency of less than a second. It is the novel consensus protocol that has opened the door to these performance improvements, he says.

ETH 2.0, like most proof-of-stake blockchains today, will rely on a “classical” consensus protocol, Sirer says. Under this model, every node has to communicate with every other node until two-thirds of the network signs off on a transaction. According to Sirer, this “all-to-all” node communication model slows down the transaction approval process.

AVA takes a different approach to node communication. A small group of randomly selected nodes first decide if a transaction is valid and then the network repeats the sampling process a number of times until it gets a supermajority of nodes that are willing to accept the transaction. This way, a given node does not have to communicate with every other node out there. As a result, more transactions can go through the network every second.

The need for “all-to-all” communication limits the number of validators these classical protocols can support, says Sirer. To compensate, many of them employ a “committee selection” mechanism to delegate voting to a small number of validators, which could lead to the concentration of power.

“We have seen that blockchains that permit only a small number of validators just become captured by a small cabal,” says Sirer. “In contrast, AVA can scale to millions of validators.”

Throughput vs. scalability

Georgios Konstantopoulos, an independent consultant and engineer focused on blockchain scalability, argues that one should be skeptical of AVA’s claims about scalability. Specifically, higher throughput does not necessarily equal higher scalability, he said in Telegram messages to The Block.

“AVA indeed can scale latency. From the user side, AVA gives you the guarantee that once you submit a transaction, it will be confirmed very fast,” he said. “Is this enough to claim that they have solved scaling? I would say no.”

In a blockchain network, each node must download and store the "state” – the status of all accounts, balances, and smart contract records on the blockchain – as well as all transaction data before it can sync the chain and participate in the validation process. The volume of transaction data each node must download limits who can join the network, because processing that large amount of data requires high-performance equipment.

Konstantopoulos argued that scaling a decentralized network, at its core, means not only maximizing throughput and minimizing latency but also allowing any node to sync the chain. For this reason, a network should not require every node to process every transaction. 

Furthermore, said Konstantopoulos, this is not an “apples to apples comparison.” AVA is starting from a clean slate, whereas to sync Ethereum’s blockchain a new node must download more than 200 gigabytes (and that pile of data is growing by 10 to 15 gigabytes per month).

Ethereum co-founder Vitalik Buterin told The Block in an email that AVA faces the same “scalability trilemma” that has haunted Ethereum for years. The idea is that public blockchain networks cannot simultaneously achieve decentralization, security, and scalability.

“The problem with this approach is exactly the same problem that I have been harping about for many years,” he said. “A network where all validators must process all transactions is a network whose capacity is limited to the capacity of one of these validators, meaning that either scalability must be sacrificed or the network becomes centralized through supernodes.”

Testnet results should also be taken with a grain of salt, argued Buterin. “Ethereum has also seen up to 1000 TPS in test environments; there are reasons why it only supports ~40 TPS on mainnet,” he said. “Those reasons have to do with the fact that the limits to the capacity of a network don't just come from the theoretical capacity that you can run in a test environment.”

The limits also come from centralization risk, limits to storage size, desire to limit how long it takes for a node to sync for the first time, etc, he said, adding that that Ethereum purposely maintains a low throughput, explicitly to alleviate the practical constraints that large data storage requirements would impose on validators.

Sirer said AVA will periodically rid the chain of unnecessary historical data – a process known as “pruning” – so that nodes will only have to download a trimmed down version of the state.

But Buterin said AVA’s pruning process is equivalent to Ethereum’s fast-sync that downloads only the recent state and not the entire history. It is an improvement, he said, but is not enough to resolve the problem.

According to Konstantopoulos, the only way to have significant performance gains is by relieving the burden of storing all transactions on every node. This is what off-chain or layer 2 schemes, and new approaches like “stateless Ethereum”, aim to accomplish.

In Buterin’s view, scalability will depend on sharding – an mechanism adopted by ETH 2.0 that splits the blockchain into different servers so a given node does not need to process all network data. Although AVA claims to have adopted sharding on the subnetwork level, Buterin contended that its blockchain is not truly sharded, and that it imposes a limit on the number of cross-shard transactions that it can support.

AVA’s approach to consensus is “novel and ingenious,” Buterin acknowledged. “At the same time I have long held the opinion that the consensus algorithm cannot make much difference to a blockchain's capacity.”