Consensus coprocessor
Dig into the consensus mechanisms behind Recall's consensus mechanisms.
In addition to scaling through hierarchical subnets, Recall employs a unique approach to consensus that allows each subnet to efficiently process large volumes of data. This approach is based on a parallel consensus coprocessor that enables the network to reach agreement on the order of data-carrying transactions, even as the data itself is being optimistically synchronized among validators in the background. This coprocessor protocol is designed to be fast, secure, and flexible.
Features
Recall's consensus coprocessor decouples execution from consensus, allowing the blockchain to process transactions and manage data in parallel, while retaining deterministic state updates and security. This coprocessor leverages five key components to achieve this:
- Multi-phase BFT coprocessor: Recall's coprocessor operates as a layer between consensus and execution, dividing state machine replication into distinct phases (e.g., proposal, pre-vote, pre-commit, and commit).
- Optimistic data sync and execution: Transactions and data synchronization tasks are optimistically executed in parallel prior to commit stage, significantly reducing latency and improving processing speed.
- Async execution off the hotpath: Execution of data-carrying transactions is handled separately from the consensus mechanism, using a distributed task pool, which allows for independent/async processing.
- Multi-phase data availability: A quorum voting process ensures data availability and integrity during consensus, supported by robust erasure coding techniques to protect against longer-term data loss.
- Distributed execution pool: Separating task execution from ordering enables continuous task processing even as data synchronization occurs asynchronously in the background.