Solana validators approved an update that aims to settle transactions in just 150 milliseconds.
Solana validators last week approved a consensus upgrade for the network that would cut transaction finality dramatically. But some experts are concerned that faster finality brings tradeoffs, including in security.
Over 98% of nodes backed the new Alpenglow upgrade, according to a Sept. 2 post from Solana Status on X. The updated consensus protocol for Solana, developed by Anza, a spinoff from Solana Labs, is expected to bring near-instant finality to transactions, reducing the current 12-second wait to just 150 milliseconds.
For Solana, Alpenglow marks one of its boldest technical moves since the network’s outages in 2022 and 2023.
If it works as intended, the upgrade will replace older systems like proof-of-history voting and gossip messaging, with a leaner design where most of the consensus runs off-chain. Rather than each validator pushing votes directly onto the chain, they’ll bundle them into compact certificates.
Muriel Médard, an award-winning professor of software science and engineering at MIT, as well as the co-founder of blockchain infra company Optimum, told The Defiant that latency reduction and the desire to be at speeds compatible with web2 lie at the heart of Alpenglow.
“This an important goal and one that per se much of the Web3 community applauds. The question is not so much whether this is a reasonable desideratum but whether the tech can actually achieve it,” Médard said.
Security Compromises
There’s no fixed timeline yet for Alpenglow, as it still needs to go through testing to verify security. If everything stays on track, Solana RPC firm Helius expects the upgrade to reach the mainnet by early 2026.
Other networks, including Coinbase’s Base, Unichain, and BNB Chain, are also pushing for faster finality. For instance, in mid-July, Base rolled out Flashblocks, a new feature that cuts block times from two seconds to 200 milliseconds.
And even though faster finality promises more efficient markets and smoother user experiences, it also raises questions about tradeoffs. Back in May, Bitcoin developer Jeff Garzik told The Defiant that achieving such speeds is impossible without security compromises.
Asked about blockchain security, he said at the time that “there’s a certain speed beyond which you literally can’t go over a fiber optic cable through the ocean to another continent and then back again within a certain number of milliseconds. And if you’re faster than that, you’re just giving up decentralization for speed. And when you do that, you’re also giving up security for speed.”
Médard, who is one of the key inventors of random linear network coding, also highlighted a limitation in the Alpenglow execution related to propagation, handled by a component of Alpenglow called Rotor. In the case of blockchain, propagation refers to the transmission and distribution of new data, aka blocks, across the network.
“Let us concentrate on the aspect of Alpenglow that the protocol generally cannot control, which is the propagation, embodied by Rotor in Alpenglow. In this part of the Alpenglow execution, the uncertainty in connections that are occurring over Web2 cannot be controlled and therefore that can seriously impact the fast finality sought by Alpenglow,” she said.
Extra Data Can Slow Network
Rotor uses a coding method that sends extra data to cover for any pieces that arrive late, Médard explained. She added that the approach Alpenglow’s Rotor takes is to use “a 1950-60s code” called Reed-Solomon, originally “designed for a different purpose and at a time when computation was paltry, placing all sorts of constraints that are currently irrelevant.”
“The instantiation of RS Rotor uses is a rate ½ code. What that means is half of the bandwidth is sacrificed to transmitting pieces that are there to make up for the pieces that were supposed to arrive on time but did not,” Médard said.
The MIT professor added further that Rotor sends extra pieces of data to cover for delays, noting that this wastes resources if the network is clear and makes things slower if the network is busy.
She added: “The redundancy/repair should be carefully tuned. Sometimes, but seldom, rate ½ will be the right redundancy for a single validator. The probability that rate is the correct one for all the validators, that the route from the supplier to each destination has exactly the same characteristics, is basically nil. The only code known to be optimal for erasures independently of the road network and its condition is random linear network coding (RLNC).”
The Defiant reached out to Alpenglow developer Anza for comment on both Médard’s and Garzik’s concerns around the consensus protocol, but has not heard back by press time.
Solana’s ecosystem has continued to grow in recent months, with total value locked (TVL) on the network breaking new highs, led by USDC issuer circle and lending protocol Kamino.