CoinYQ Dossier

A clock built for speed learns how to recover

Solana did not begin with a faster virtual machine. It began with Anatoly Yakovenko’s claim that a blockchain could waste less time agreeing about time. Greg Fitzgerald, Stephen Akridge and Raj Gokal turned that premise into a Rust network that launched in March 2020. Its speed attracted exchanges, traders and FTX-linked capital; its tightly coupled machinery also made every halt a public stress test. By 2026 the story had shifted from one fast client to several implementations. That transition matters because throughput, uninterrupted service and independent software are three different achievements.

Yakovenko turns a late-night clock idea into a team

In 2017 Anatoly Yakovenko sketched a ledger whose SHA-256 sequence could prove that time had elapsed and place messages in order. The white paper carried his name, but the working network emerged from divided roles: Raj Gokal joined in December 2017 to build the company, and former Qualcomm colleagues Greg Fitzgerald and Stephen Akridge arrived in early 2018.

Fitzgerald persuaded Yakovenko to replace the C prototype with Rust and released an early open-source implementation. Akridge argued that GPUs could verify signatures in parallel. Their company briefly used the name Loom, then chose Solana after the California beach where the Qualcomm colleagues had surfed.

Proof of History supplies the clock; validators still decide

Proof of History repeatedly hashes one output into the next. Because producing the sequence is sequential but checking recorded samples can be parallelized, a leader can insert transaction identifiers into an order other nodes can verify. It tells the network that event A came before event B and that computation occurred between them.

It does not vote a fork into truth. SOL-backed validators replay blocks and cast stake-weighted votes. Tower BFT uses the PoH clock to enforce increasing lockouts, making it costly for a validator to reverse an earlier vote. Mainnet Beta began on March 16, 2020 with this combination of clock, leader schedule, parallel runtime and proof-of-stake consensus.

Benchmarks met the harder metric: staying live

Before launch, Solana reported more than 50,000 transactions per second on a GPU-equipped testnet of over 200 physically distinct nodes. The qualifiers are part of the fact. It was a 2019 testnet benchmark, and later RPC interfaces would explicitly separate all transactions from non-vote transactions over a sample window.

On September 14, 2021, bots around the Grape Protocol sale flooded the network. Memory exhaustion crashed validators, consensus stopped and the cluster remained offline for 17 hours. Operators coordinated a restart; the Foundation reported that no funds were lost. The recovery demonstrated social and operational coordination, while the halt showed that high peak capacity did not guarantee liveness under pathological load.

The next year brought different failures rather than one reusable diagnosis. April’s seven-hour outage followed roughly six million inbound transactions per second—attempted traffic, not completed user TPS. June exposed a durable-nonce bug. September combined duplicate blocks with a fork-selection edge case. In February 2024 a LoadedPrograms cache loop stopped finalization for about five hours; engineers shipped v1.17.20 and validators restarted from an agreed slot.

FTX bought into the ecosystem, then disappeared without stopping the chain

The Foundation’s November 2022 fact sheet documents 58,024,833 SOL in settled sales to FTX/Alameda: Foundation tranches of 4 million, 12 million and 34,524,833 SOL, plus a 7.5 million-SOL Solana Labs sale to Alameda Research Ventures. Another Labs sale of 61,853 SOL was marked not yet settled. The 12 million tranche was scheduled for monthly unlocks through September 2027, the 34,524,833 tranche was scheduled through January 2028, and the 7.5 million tranche was scheduled to unlock on March 1, 2025. Those dates made the bankruptcy estate’s exposure a supply question as well as a reputational one.

The Foundation separately disclosed about US$1 million of cash or equivalents trapped on FTX.com, no SOL custodied there, and holdings of roughly 3.24 million FTX shares, 3.43 million FTT and 134.54 million SRM. Those losses and token sales explain the ecosystem’s exposure without turning FTX into the operator of Solana consensus. The chain did not suffer a notable uptime or security incident during the bankruptcy week; projects, liquidity and confidence absorbed the shock instead.

The original client leaves Solana Labs and gains competitors

The outages made software lineage a security question. In 2024 former Solana Labs engineers formed Anza and forked the original validator into Agave. Governance of the repository moved to a new firm, but Agave remained a continuation of the same code family; an organizational split alone could not remove common-mode defects.

Independent implementations were the deeper change. The August 13, 2026 Foundation changelog listed Agave v4.2.0, full Firedancer Mainnet v1.1.4 and hybrid Frankendancer Mainnet v0.1105.40200. The August 28 update then listed Firedancer Mainnet v26.08.2, Agave v4.3.0-beta.2 and a newer Frankendancer testnet build. Firedancer documentation explains that Frankendancer retains Agave components. These versions prove active alternatives, not equal stake adoption; new clients also bring their own bugs and operating work.

SOL links speed to validator economics

Every transaction pays SOL. Current documentation sets a 5,000-lamport base fee per signature, half burned and half paid to the block producer, plus an optional priority fee paid to that validator. Delegated SOL weights consensus votes and receives inflationary rewards after commission.

The published inflation path began at 8% annually, declines 15% each year and approaches 1.5%. That schedule is not a promised staking yield: the return also depends on how much SOL is active, validator uptime and commission. Solana’s engineering story therefore ends where it runs each day—with operators deciding which client to trust and holders deciding which validator receives their stake.

How the project changed

  1. 2017
    Yakovenko writes the clock design

    A sequential SHA-256 record becomes the basis for ordering events before consensus communication.

  2. 2017-12
    Gokal joins the project

    Raj Gokal adds the company-building role to Yakovenko’s protocol experiment.

  3. 2018-Q1
    Fitzgerald and Akridge turn the paper into software

    The implementation moves to Rust and GPU signature verification becomes part of the performance plan.

  4. 2020-03-16
    Mainnet Beta begins

    The genesis block launches the production cluster with PoH, Tower BFT and stake-weighted validators.

  5. 2021-09-14
    Bot traffic precedes a 17-hour halt

    Memory exhaustion crashes validators; operators coordinate a restart and the Foundation reports no lost funds.

  6. 2022-04—2022-09
    Three outages expose three failure modes

    Inbound load, durable-nonce handling and a fork-selection bug each stop block production and lead to separate fixes.

  7. 2022-11
    FTX bankruptcy reveals the commercial exposure

    The Foundation publishes SOL sale schedules and its FTX, FTT and SRM holdings while consensus continues operating.

  8. 2024-02-06
    A program-cache loop stops finalization

    Engineers cut v1.17.20 and validators restart from an agreed slot after roughly five hours.

  9. 2024-03
    Agave forks from the Solana Labs client

    Anza takes over the original code lineage as a new engineering organization.

  10. 2026-08-28
    Independent and hybrid client lines keep shipping

    The latest changelog lists Firedancer Mainnet v26.08.2 after an earlier August update recorded Mainnet releases for Agave, full Firedancer and Frankendancer.

Evidence and primary sources

Last evidence review: 2026-09-05

More stories about this project

What is Solana?

Solana is a single-state proof-of-stake blockchain launched as Mainnet Beta on March 16, 2020. Proof of History records a verifiable sequence that helps nodes agree on when events occurred, while Tower BFT and stake-weighted validator votes choose among forks. SOL pays transaction fees, can be delegated to validators and receives protocol inflation rewards. Applications execute in parallel when their account access does not conflict, which is one reason the network can process many transactions without splitting state across shards.

What problem does Solana solve?

Anatoly Yakovenko wanted a distributed network to approach the performance of one machine. He treated coordination about time as the bottleneck: if validators could verify a shared sequence locally, they could spend less communication deciding order. Greg Fitzgerald moved the prototype to Rust, Stephen Akridge pushed signature checks onto GPUs, and Raj Gokal helped turn the engineering experiment into a company. The same tightly integrated design later made liveness failures highly visible. A fast chain that stops cannot settle anything, so each outage became a test of both software and validator coordination.

How does Solana work?

A scheduled leader receives transactions, orders them against the Proof of History sequence and executes non-conflicting work in parallel. Other validators replay the block and cast stake-weighted votes; Tower BFT uses those votes and the PoH clock to lock out conflicting forks. Users pay fees in SOL, and delegators assign stake to validators whose votes help secure the ledger. Current RPC samples expose both total transactions and non-vote transactions over a stated number of seconds, so a TPS number that omits vote traffic and its sample window is incomplete. The 2019 claim above 50,000 TPS was a GPU-equipped testnet benchmark with more than 200 nodes, while the six million per second reported in April 2022 was inbound traffic during an outage, not successful user throughput.

Key facts

  • Yakovenko conceived the sequential-hash clock in 2017; Gokal joined in December 2017, and Fitzgerald and Akridge joined in early 2018 with distinct software and hardware-acceleration roles.
  • Solana created its genesis block and used it to launch Mainnet Beta on 2020-03-16.
  • Proof of History records event order and verifiable passage of computation; Tower BFT and stake-weighted proof-of-stake votes perform consensus.
  • The often-cited 50,000-plus TPS result came from a 2019 GPU testnet with more than 200 physically distinct nodes, not a guaranteed Mainnet rate.
  • The 17-hour 2021-09-14 outage followed bot traffic, memory exhaustion and validator crashes; validators restored the cluster through a coordinated restart.
  • The April, June and September 2022 outages had different causes: inbound load, a durable-nonce bug, and duplicate blocks plus a fork-choice bug.
  • The 2024-02-06 outage lasted about five hours; v1.17.20 removed the legacy loader path implicated in the LoadedPrograms loop before the restart.
  • The November 2022 table documented 58,024,833 SOL in settled sales to FTX/Alameda: 50,524,833 from the Foundation and 7,500,000 from Solana Labs. Another 61,853 SOL was then marked unsettled.
  • The published inflation schedule starts at 8%, falls 15% year over year and approaches 1.5%; staking yield is a different, variable number.
  • In August 2026 official records listed Mainnet releases for Agave, full Firedancer and hybrid Frankendancer; the August 28 update named Firedancer Mainnet v26.08.2, but release availability does not prove equal validator adoption.

Official links

Categories

Related coins

Frequently asked questions

Who created Solana?

Anatoly Yakovenko wrote the original clock-based design. Raj Gokal joined in December 2017; former Qualcomm colleagues Greg Fitzgerald and Stephen Akridge joined in early 2018. Fitzgerald shifted the implementation from C to Rust, while Akridge proposed GPU acceleration for signature verification.

Is Proof of History Solana’s consensus mechanism?

Not by itself. Proof of History supplies a verifiable ordering clock. Validators stake SOL and cast votes, and Tower BFT uses those stake-weighted votes and the clock to decide forks and finality.

Does Solana process 50,000 user transactions every second?

That number came from a 2019 GPU testnet with more than 200 nodes. Live measurements vary, and Solana reports total transactions and non-vote transactions separately. A credible TPS figure must state the date, sample duration and whether validator votes are counted.

Why has Solana stopped producing blocks?

There was no single recurring cause. The 2021 halt involved bot load, memory exhaustion and validator crashes; 2022 incidents involved inbound load, durable nonces and fork selection; the 2024 halt came from a program-cache loop. Recovery required validators to agree on a restart point and run patched software.

Did FTX own or control Solana?

The documented tie was substantial but commercial. FTX/Alameda bought tens of millions of SOL from the Foundation and Solana Labs and supported ecosystem projects. The Foundation also held FTX-related assets. Its bankruptcy did not take down consensus, and holding SOL does not let an entity move stake whose keys it does not control.

What changed with Agave and Firedancer?

Agave moved stewardship of the original Solana Labs validator lineage to Anza. Firedancer is a separate implementation, while Frankendancer retains Agave components. Official August 2026 updates listed Mainnet releases for all three lines and later named Firedancer Mainnet v26.08.2; those facts do not show that stake is evenly distributed among them.

What is SOL used for?

SOL pays base and optional priority fees and can be delegated to validators. Delegators may receive inflation rewards after validator commission, but the return changes with inflation, active stake and validator performance.

External trackers

Choose a tracking site for Solana: