How Solana works
Accounts, programs, parallel execution, leaders and local fee markets. The architecture that makes 200ms blocks possible.
Solana was designed around one idea: make a single global state machine as fast as the hardware and network allow, without splitting it into shards or layers. A few design choices make that possible.
Accounts and programs
Everything is an account: a chunk of data with an owner, a SOL balance and an address. Programs (smart contracts) are accounts too, but they're stateless: they hold code, not data. A lending market's pools, your positions and your token balances all live in separate accounts that the program owns and edits.
Every transaction lists, up front, every account it will read or write. That one rule unlocks most of Solana's speed.
Sealevel: parallel execution
Because each transaction declares its accounts, the runtime (called Sealevel) can see which transactions don't overlap and run them at the same time across CPU cores. A swap on Orca and a deposit on Kamino touch different accounts, so they execute in parallel. Two swaps in the same pool must run one after the other.
Leaders, slots and Turbine
Validators take turns as leader for a few consecutive slots, on a schedule published in advance. Wallets and RPCs forward transactions straight to the upcoming leaders instead of a global waiting room (a mempool), which cuts latency.
Since October 2026, slots are 200 milliseconds (down from 400). The leader streams its block out in small pieces called shreds, fanned out through a tree of validators by a protocol called Turbine, so no single node has to send the whole block to everyone. Alpenglow's planned Rotor will eventually replace Turbine with a simpler single-hop relay.
Fees: base, priority and local markets
Every transaction pays a base fee of 5,000 lamports per signature. During busy periods, a priority fee (priced per unit of compute) helps your transaction get scheduled sooner. Because transactions compete only with others touching the same accounts, fee markets are local: a frenzied memecoin launch makes that token's pool expensive without raising fees for someone lending USDC.
Rent and state
Storing data costs a refundable rent deposit proportional to account size, which discourages bloating the chain with abandoned accounts. Close an account and the deposit returns.
Clients
Validators run software called a client. Agave (by Anza) is the main one; Jito-Solana adds the MEV block engine; Firedancer (by Jump) is an independent rewrite live on mainnet since December 2025. Multiple clients mean a bug in one doesn't take down the network. More in Validators and SOL economics.
Key takeaways
- Programs are stateless; all state lives in accounts that transactions must declare up front.
- Declaring accounts lets Sealevel run non-overlapping transactions in parallel.
- A known leader schedule lets transactions skip a global mempool.
- Fee markets are local: congestion on one hot account doesn't make everything expensive.
Check yourself
3 questions. Get them all right to mark this lesson complete.
Try it
Facts checked Oct 19, 2026. Educational content, not financial advice.


