Raptr: The Consensus Layer for the Global Trading Engine

Development
,
1

Authors: Andrei Tonkikh, Balaji Arun, Zhuolun Xiang, Zekun Li, Alexander Spiegelman

:white_check_mark: Core Idea: Prefix Consensus

Raptr introduces Prefix Consensus, a new paradigm that makes leader-based BFT protocols faster, more robust, and scalable by combining the strengths of optimistic and pessimistic approaches.

It achieves the elusive BFT trifecta:

  • Extremely high throughput (up to 250,000 TPS)
  • Near-optimal latency (~750ms)
  • Decentralized robustness, even under network glitches.

:building_construction: How It Works:

  • Instead of voting for the entire proposed block or abstaining, replicas vote on the longest prefix they’ve received(partial voting).
  • Consensus proceeds even if some nodes miss part of the data—avoiding blocking on slow/faulty nodes.
  • Availability and safety quorums are decoupled:
    • Only f+1 votes needed to certify a prefix as available
    • Remaining votes complete the safety quorum (2f+1)
  • The next leader extends either the full-prefix QC or the longest certified prefix.

This approach ensures:

  • Safety (committed prefixes are always contained in future blocks)
  • Progress at network speed, tolerating partial data loss or delays.

:rocket: Why It’s Better:

:white_check_mark: Parallel Data Dissemination like Narwhal/Bullshark removes leader bottlenecks.
:white_check_mark: Faster than pessimistic designs (e.g., Jolteon + Quorum Store) by cutting down message delays.
:white_check_mark: More robust than optimistic DAG-based designs (e.g., Mysticeti), which suffer large latency spikes under minor network issues.

:bar_chart: Performance:

  • 260,000 TPS with ~750ms latency in geo-distributed tests (100 nodes).
  • Maintains stable throughput & latency even under 1% message drop, unlike DAG-based protocols that degrade up to 10x latency.

:trophy: Key Innovations:

  1. Prefix voting + partial commits (vote/commit on available prefixes, not all-or-nothing)
  2. Decoupling availability from safety quorums
  3. Separate network channels for consensus vs. data
  4. Reputation-based leader & batch selection
  5. Mitigation for triangle inequality (delayed inclusion of new batches to ensure propagation)

:memo: In Summary:

Raptr’s Prefix Consensus represents a paradigm shift in BFT protocol design—bridging speed, scalability, and robustness. It overcomes tradeoffs in existing designs by allowing consensus to proceed with partial data, decoupling availability from safety, and enabling higher throughput without sacrificing fault tolerance.

Read more: