1) Why we need sharding: if blockchain consensus relies on every node processing all transactions + storing all state, the network cannot scale beyond the capacity of a single node.
2) So need to achieve consensus on subsets of transactions (the raw data) and state (the results of interpreting the data) in a way that balances security and availability.
3) The current design has four layers: 1) a main chain contract, responsible for validator deposits and slashing conditions, 2) a random beacon chain, 3) consensus on data (transactions), 4) consensus on state.
4) Core idea: use the random beacon to enlist subsets of the global validator pool for service across shards; regularly shuffle them to lower probability of collusion and bribery.
5) Another core idea: strike a balance between availability and security by escalating validation requirements from local to global via checkpoints – attestation, notarization, meta-notarization.
6) Other elegant constructions in the design – e.g. "cryptoeconomic aggregated signatures”: use a bitfield to summarize signatures and rely on the profit-motive of signers to challenge wrong data.
7) Cool to see ideas from other projects also inspiring the design – open-source in action – Dfinity’s random beacon, Truebit verification.
8) The system will enable future upgradability of the network. Since rules are shard-specific: a new shard can have a different VM (e.g. WASM), a different inter-contract comms protocol (e.g. Primea), etc.
9) The architecture makes sense but there are details to figure out –– e.g. how to lessen load on the mainchain, how to use Proof of Custody within aggregate signatures, … with Ethereum grant funding available and sharding important work, it’s a good time to contribute.
10) The implementation roadmap takes on battles one at a time: build 1) the random beacon, 2) the data availability solution, then 3) layer on state.
11) What’s heartening: the system will be useful at various degrees of completion. E.g. the data layer, with no interpretation built in, can be used as a filesystem for Truebit tasks.
12) For more details, watch this awesome talk by Justin Drake at Edcon:
13) Thanks to @drakefjustin , @karl_dot_tech , @VitalikButerin , @icebearhww , @jon_choi_ , @davidlknott for discussions – super interesting & important area of work!
