Solana takes its votes off the chain
Solana's Alpenglow upgrade went live on Wednesday, Anza said, moving validator votes off the blockchain and targeting finality of about 150 milliseconds. Validators now pay a flat 0.8 SOL ticket each epoch instead of per-vote fees.
Solana's Alpenglow upgrade went live on Wednesday, moving validator votes off the blockchain and targeting transaction finality of about 150 milliseconds, according to Anza, the developer that built it. Anza said the new consensus protocol reached mainnet-beta at slot 459,653,000 at about 14:00 UTC, The Block reported. Finality is the point after which a transaction cannot be reversed.
The switch came early in epoch 1064. By our reading of a public Solana node, that epoch began at slot 459,648,000 at 13:36 UTC, so the handover came 5,000 slots later. At 15:44 UTC the node listed 666 vote accounts as current and 11 as delinquent. Brennan Watt, Anza's chief executive, called it "the largest, most ambitious blockchain upgrade since 'The Merge'", Ethereum's 2022 switch to proof of stake, according to The Block.
The two accounts of the starting point differ. Anza put finality before the upgrade at 12.8 seconds, The Block reported, while CoinDesk, writing before the switch, said it took about six seconds and that the new target would be roughly 40 times faster. The 150 millisecond figure is Anza's target; The Block reported no measured figure after the switch.
The mechanism changes how agreement is recorded. Votes are now signed with BLS signatures, a scheme that lets many signatures be combined into one, and aggregated into compact certificates rather than written to the chain as ordinary transactions, Galaxy said in a report on its experience running a validator on the test network. Proof of History, Solana's built-in clock, retires once the migration completes. Slashing for double voting is "well defined but deliberately not enabled in the first iteration", Galaxy wrote.
The fault model shifts as well. Galaxy said Alpenglow is built for "20 plus 20": up to 20 per cent of stake acting maliciously plus a further 20 per cent offline. The previous protocol, TowerBFT, tolerated up to 33 per cent malicious stake. Galaxy, one of the larger Solana validator operators, sells staking services, and its account is its own.
Costs for validators change too. Galaxy estimated that voting cost about 1 SOL a day at 400 millisecond slots and about 2 SOL a day at the current 200 milliseconds; Alpenglow replaces that with a flat validator admission ticket of 0.8 SOL per epoch, which runs about a day. By our arithmetic that is 60 per cent less than 2 SOL a day, the saving CoinDesk also gave. A vote account that cannot cover the ticket at an epoch boundary "is excluded from consensus for that entire epoch", Galaxy said.
The test network was uneven. It launched in early May with around 50 independent operators and grew to roughly 100, Galaxy said, and was restarted from a fresh genesis block six times. Galaxy describes a migration that failed network-wide and a rollback, though its text is not consistent on the order of events. Validators were asked to run the Agave client for the switch; Firedancer's 26.10 release supports Alpenglow, and the Frankendancer hybrid will not be supported now that it is active.
The validator base has thinned. CoinDesk, citing Bitwise's staking report, said active validators fell to 659 at the end of September from 696 three months earlier, about 30 per cent fewer than a year before; by our subtraction that is 37 fewer in a quarter. Citing Solana Compass data, it said 69 per cent of SOL's supply was staked, that the smallest group of validators controlling more than a third of stake fell to 18 from 20 over the past year, and that the top ten's share rose to 24.5 per cent from 22.2 per cent.
Robin Nordnes, chief executive and founder of Raiku, which builds trading infrastructure on Solana, told CoinDesk: "Most of what got counted before was validators voting, and that's off-chain now." His point is that vote transactions, which made up most of Solana's counted activity, no longer appear on chain, so activity figures from before and after the switch are not like for like.
Why it matters: a flat, lower cost of taking part should help smaller operators, but the concentration figures show where the network starts from, and the first epochs will show how many vote accounts clear the new ticket. Komposite has looked at the economics of running validators in institutional staking, and at how regulators now classify the assets on top, in the SEC's March interpretation.